KR20210061826A - Refrigerator operating method - Google Patents

Abstract

Disclosed is a freezer operating method which is able to use an artificial intelligence (AI) algorithm and/or machine learning algorithm in a 5G environment connected for the Internet of things (IoT). According to an embodiment of the present invention, the freezer operating method can comprise: a step of receiving an input of operation data on an operation status of a freezer; a step of determining whether a compressor placed in the freezer is in a normal status, a status with an impending surge, or a status with a surge based on the operation data; a step of, when the compressor is in the status with an impending surge or the status with a surge, controlling to block the surge; and a step of, when the compressor is in the status with an impending surge or the status with a surge, outputting a notification information. The determination of the presence or absence of a surge may be to determine whether at least one factor included in the operation data is located in which of a normal area, an area with an impending surge, or an area with a surge of the comparison data. The comparison data may be drawn by learning in accordance with the AI model using a support vector machine.

Description

Refrigerator operation method {REFRIGERATOR OPERATING METHOD}

The embodiment relates to a method of operating a refrigerator, and more particularly, to a method of operating a refrigerator including a process of determining whether a surge has occurred in a compressor provided in the refrigerator.

The content described in this section merely provides background information on the embodiment and does not constitute the prior art.

A refrigerator is a device that transfers heat from a low-temperature thermal reservoir to a high-temperature heat storage tank, and is widely used for home and industrial use. In order to satisfy the request of consumers and for efficient operation, the refrigerator is continuously being improved, and various studies are being conducted for the improvement of the refrigerator.

The refrigerator is equipped with a compressor for compressing the refrigerant. Surge may occur in the compressor. Surge occurs when the compression ratio of the compressor is high compared to the flow rate of the refrigerant, and refers to a phenomenon in which the flow of the refrigerant flow becomes irregular as the rotating body of the compressor is idle.

In the event of such a surge phenomenon, the compressor cannot produce a pressure greater than the pressure resistance of the refrigerator. Accordingly, when a surge occurs, reverse flow of the refrigerant may occur repeatedly and damage to the compressor may occur.

Therefore, it is necessary to detect the occurrence of such a surge or predict the occurrence of the surge to prevent damage to the compressor due to the surge.

In the embodiment, a method of detecting or predicting the occurrence of a surge in a compressor is proposed.

In the embodiment, a method of operating a refrigerator is proposed in order to prevent a surge from occurring before the surge occurs according to a predicted result, and to minimize damage to a compressor and a refrigerator caused by the surge once it occurs.

In the embodiment, a method of predicting surge occurrence in a compressor through artificial intelligence model learning is proposed using operation data that can be obtained from a normal operation of a refrigerator.

The technical problem to be achieved by the embodiment is not limited to the technical problem mentioned above, and other technical problems that are not mentioned will be clearly understood by those of ordinary skill in the technical field to which the embodiment belongs from the following description.

In order to achieve the above-described problem, the method of operating a refrigerator according to an embodiment includes the step of receiving operation data related to the operation state of the refrigerator, the compressor provided in the refrigerator from the operation data is in a steady state, a surge is imminent, or The step of determining whether a surge is generated to determine whether a surge is in a state of occurrence, the step of controlling to block the occurrence of a surge when the compressor is in an impending surge or a state of generating a surge, and the compressor is in a state of imminent surge occurrence or a state of generating a surge If yes, it may include the step of outputting the notification information.

The determination of whether or not the surge is generated may be to determine whether at least one factor included in the operation data is located in a normal region, an imminent surge generation region, or a surge generation region of the comparison data.

The comparison data may be derived by learning according to an artificial intelligence model using a support vector machine.

The support vector machine may set a boundary between a normal region, an imminent surge occurrence region, or a surge occurrence region of the comparison data.

The refrigerator includes a compressor, a condenser in which the inlet is connected to the outlet of the compressor, an expansion device in which the inlet is connected to the outlet of the condenser, the evaporator in which the inlet is connected to the outlet of the expansion device and the outlet is connected to the compressor inlet, and the operation of the refrigerator. It may include a control unit.

The control unit is connected to a processor that derives comparison data, and the processor performs learning according to an artificial intelligence model using operation data, but may derive comparison data corresponding to each cooling capacity of the refrigerator. .

In the AI model learning, the steps of selecting factors that are the criteria for determining whether surges occur in the driving data, deriving labeling data from the selected factors, and normalizing training data provided with factors selected from the driving data It may include a data pre-processing step including a process of, optimizing an artificial intelligence model using a support vector machine, and performing learning according to an optimized artificial intelligence model using training data.

The selection of factors can be to determine the order of the factors in the order in which the frequency of surge occurrence is high as the numerical value changes, and select the number of factors that are set sequentially from the highest priority factors. have.

The derivation of the labeling data may be to select a part of the operation data and classify the selected operation data into one of a normal region, an impending surge region, or a surge occurrence region.

The learning may be performed based on the labeling data, and may set a boundary between a normal region, an imminent surge occurrence region, or a surge occurrence region in the training data.

The learning may be performed by setting a boundary of any one of a normal region, an imminent surge occurrence region, or a surge occurrence region in the training data, and then sequentially setting the boundaries of the remaining two regions.

The comparison data may be derived from the training data in a state in which the boundary of a normal region, an imminent surge occurrence region, or a surge occurrence region is set, and the determination of whether a surge occurs may be performed at a set number of times within a set time.

Data pre-processing may include deleting missing values from the training data.

Optimization of the artificial intelligence model may include selecting a hyperparameter.

The surge generation impending state may be defined as a state from a time point before a set time from a surge occurrence time point of the compressor to the surge occurrence time point.

The factor may include at least one of a current applied to the compressor or a refrigerant temperature at an inlet of the compressor.

In the embodiment, since it is possible to determine about the occurrence of a compressor surge by learning an artificial intelligence model, it is possible to significantly reduce the occurrence of a judgment error compared to determining the occurrence of a surge by another device.

In the embodiment, since it is possible to determine the occurrence of a compressor surge using only information obtained from various sensors already provided in the refrigerator without using a separate device, there is an effect of reducing the production cost of the refrigerator.

In an embodiment, by removing missing values from training data used for learning an artificial intelligence model, it is possible to shorten the learning time and increase the accuracy of the learning result.

1 is a view for explaining a refrigerator according to an embodiment.
2 is a flowchart illustrating a method of operating a refrigerator according to an exemplary embodiment.
3 is a flow chart illustrating learning an artificial intelligence model according to an embodiment.
4 is a graph showing the distribution of factors related to a surge in a compressor in learning an artificial intelligence model according to an embodiment.
FIG. 5 is a diagram illustrating a boundary for dividing a surge generation region and a remaining region in the graph of FIG. 4.
6 is a diagram illustrating a boundary for dividing an area that is imminent to generate a surge and the remaining areas in the graph of FIG. 4.
7 is a diagram illustrating a boundary dividing a normal region from a remaining region in the graph of FIG. 4.
8 is a flowchart illustrating determination of a surge of a compressor in a refrigerator according to an exemplary embodiment.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings. Since the embodiments can be modified in various ways and have various forms, specific embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the embodiment to a specific form of disclosure, it should be understood to include all changes, equivalents, or substitutes included in the spirit and scope of the embodiment.

Terms such as "first" and "second" may be used to describe various elements, but the elements should not be limited by the terms. The terms are used for the purpose of distinguishing one component from another component. In addition, terms specifically defined in consideration of the configuration and operation of the embodiment are only for describing the embodiment, and do not limit the scope of the embodiment.

In the description of the embodiment, in the case of being described as being formed on "upper (top)" or "bottom (on or under)" of each element, upper (upper) or lower (lower) (on or under) ) Includes both elements in direct contact with each other or in which one or more other elements are indirectly formed between the two elements. In addition, when expressed as “up (up)” or “on or under”, the meaning of not only an upward direction but also a downward direction based on one element may be included.

In addition, relational terms such as "top/top/top" and "bottom/bottom/bottom" used below do not necessarily require or imply any physical or logical relationship or order between such entities or elements, It may be used to distinguish one entity or element from another entity or element.

1 is a view for explaining a refrigerator according to an embodiment. In an embodiment, a refrigerator is a device that transfers heat from a low-temperature thermal reservoir to a high-temperature heat storage tank, and includes a freezer, a refrigerator in a narrow sense, and a heat pump. , An air conditioner, and the like, and hereinafter, it is collectively referred to as a refrigerator for clarity.

The refrigerator of the embodiment may implement a refrigeration cycle using, for example, a refrigerant undergoing a phase change.

The refrigerator may include a compressor 110, a condenser 120, an expansion device 130, an evaporator 140, and a control unit 150.

The compressor 110 may have an outlet connected to the condenser 120. The condenser 120 may have an inlet connected to the outlet of the compressor 110. The expansion device 130 may have an inlet connected to the outlet of the condenser 120. In addition, the evaporator 140 may have an inlet connected to an outlet of the expansion device 130 and an outlet connected to an inlet of the compressor 110.

In this way, the compressor 110, the condenser 120, the expansion device 130, and the evaporator 140 are connected to each other through a pipe, and the refrigerant is transferred to the compressor 110, the condenser 120, the expansion device 130, The evaporator 140 may circulate and flow.

The compressor 110 compresses a refrigerant, mostly in a gaseous state, and discharges it at high temperature and high pressure. The refrigerant discharged from the compressor 110 may flow into the condenser 120.

The condenser 120 is connected to the compressor 110 through a pipe, and the refrigerant flowing into the condenser 120 is discharged into a high-temperature heat storage tank and condensed into a liquid, and the refrigerant is mostly liquid. 120).

The expansion device 130 is connected to the condenser 120 by a pipe, and expands or throttling the refrigerant flowing from the condenser 120 to make the temperature and pressure fall in a saturated state.

The evaporator 140 is connected to the expansion device 130 by a pipe, and at least a part of its surface is disposed in a low-temperature heat storage tank, so that the refrigerant flowing inside the evaporator 140 evaporates and absorbs heat from the low-temperature heat storage tank. can do.

The low-temperature refrigerant discharged from the expansion device 130 absorbs heat in the evaporator 140 and vaporizes, that is, evaporation proceeds, and is discharged from the evaporator 140 in a state in which most of it is gas, and then flows back into the compressor 110. I can.

The refrigerator may include a hot gas line 180 and a hot gas valve 181. The hot gas line 180 may be provided as a pipe in which one side is connected to the outlet of the compressor 110 and the other side is connected to the outlet of the evaporator 140.

The refrigerant discharged to the outlet of the compressor 110 may be introduced into the evaporator 140 by bypassing the condenser 120 and the expansion device 130 through the hot gas line 180. The hot gas line 180 may be provided for precise temperature control of the refrigerant in the condenser 120 and the evaporator 140.

The hot gas valve 181 may be disposed on the hot gas line 180 and electrically connected to the controller 150. The controller 150 may control opening and closing of the hot gas valve 181. The controller 150 may precisely control the temperature of the refrigerant in the condenser 120 and the evaporator 140 by controlling the flow rate of the refrigerant flowing to the hot gas line 180 through the hot gas valve 181.

The control unit 150 may be provided to control the operation of the refrigerator. That is, the control unit 150 may be electrically connected to the compressor 110, the condenser 120, the expansion device 130, the evaporator 140, and the hot gas valve 181 and control their operation.

Meanwhile, the controller 150 may be connected to the processor 160. The processor 160 may perform learning according to the artificial intelligence model to derive comparison data described below. The artificial intelligence model training will be described in detail below.

The refrigerator further includes a communication unit for communication with the server, and the control unit 150 may communicate with the server through the communication unit.

The server can store artificial intelligence models, and can also store data necessary for learning artificial intelligence models. In addition, the server can evaluate the AI model, and even after the evaluation, it can update the AI model for better performance.

The communication unit may be configured to include at least one of a mobile communication module and a wireless Internet module. In addition, the communication unit may further include a short-range communication module.

The mobile communication module includes technical standards or communication methods for mobile communication (for example, GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), CDMA2000 (Code Division Multi Access 2000)), EV-DO ( Enhanced Voice-Data Optimized or Enhanced Voice-Data Only, WCDMA (Wideband CDMA), HSDPA (High Speed Downlink Packet Access), HSUPA (High Speed Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term) Evolution-Advanced), 5G mobile communication, etc.), transmits and receives radio signals with at least one of a base station, an external terminal, and a server.

The wireless Internet module refers to a module for wireless Internet access and may be provided in a refrigerator. The wireless Internet module is configured to transmit and receive wireless signals in a communication network according to wireless Internet technologies.

The refrigerator can transmit and receive data with servers and various communication-capable terminals through 5G networks. In particular, the refrigerator communicates data with servers and terminals using at least one service from among Mobile Broadband (eMBB), URLLC (Ultra-reliable and low latency communications), and mMTC (Massive Machine-type communications) over a 5G network. can do.

eMBB (Enhanced Mobile Broadband) is a mobile broadband service, through which multimedia contents and wireless data access are provided. In addition, more advanced mobile services such as hot spots and broadband coverage for accommodating explosively increasing mobile traffic may be provided through eMBB. Through the hotspot, user mobility is small and large-capacity traffic can be accommodated in a dense area. A wide and stable wireless environment and user mobility can be guaranteed through broadband coverage.

The URLLC (Ultra-reliable and low latency communications) service defines much more stringent requirements than the existing LTE in terms of data transmission and reception reliability and transmission delay, and automation of industrial production processes, telemedicine, remote surgery, transportation, safety, etc. This is the 5G service for customers.

Massive machine-type communications (mMTC) is a service that is not sensitive to transmission delays that require a relatively small amount of data to be transmitted. A much larger number of terminals, such as sensors, etc., can simultaneously access the wireless access network by mMTC.

Surge may occur in the compressor 110. Surge occurs when the compression ratio of the compressor 110 is high compared to the flow rate of the refrigerant, and refers to a phenomenon in which the flow of the refrigerant flow becomes irregular due to the rotating body of the compressor 110 being idle.

When such a surge phenomenon occurs, the compressor 110 cannot produce a pressure greater than the pressure resistance of the refrigerator. Accordingly, when a surge occurs, reverse flow of the refrigerant may occur repeatedly and damage to the compressor 110 may occur.

Therefore, in the embodiment, the occurrence of a surge in the compressor 110 is effectively predicted, and according to the predicted result, the surge does not occur before the surge occurs, and once the surge occurs, the damage to the compressor 110 and the refrigerator caused by the surge is minimized. We propose a freezer operation method for

In an embodiment, it is possible to predict the occurrence of a surge in the compressor 110 by using operation data that can be obtained in a normal operation of a refrigerator.

At this time, the operation data is, for example, a current applied to the compressor 110, a temperature at each component of the refrigerator, a pressure ratio of the compressor 110, that is, a value obtained by comparing the pressure at the inlet and the outlet of the compressor 110, and hot The gas valve 181 may mean an open rate or other values of various factors that can be obtained from a refrigerator.

In the embodiment, the operation of the refrigerator may be performed by the control unit 150. The control unit 150 may determine the occurrence of a surge in the compressor 110 by comparing the received comparison data including the operation data of the refrigerator.

The controller 150 may determine that the refrigerator is in any one of a normal state in which a surge is unlikely to occur, an impending surge in which a surge is likely to occur soon, or a surge in which a surge has already occurred.

The comparison data, which is the criterion for determination of the control unit 150, may be divided into a normal region, an imminent surge occurrence region, or a surge occurrence region and hold the values of various factors that can be obtained from the refrigerator. Such comparative data can be derived by learning according to an artificial intelligence model. Hereinafter, the artificial intelligence model will be described.

Artificial intelligence (AI) is a branch of computer science and information technology that studies how computers can do the thinking, learning, and self-development that human intelligence can do. It means being able to imitate behavior.

In addition, artificial intelligence does not exist by itself, but is directly or indirectly related to other fields of computer science. In particular, in modern times, attempts to introduce artificial intelligence elements in various fields of information technology and use them to solve problems in the field are being made very actively.

Machine learning is a branch of artificial intelligence, a field of research that gives computers the ability to learn without explicit programming.

Specifically, machine learning can be said to be a technology that studies and builds a system that learns based on empirical data, performs prediction, and improves its own performance, and algorithms for it. Rather than executing strictly defined static program instructions, machine learning algorithms build specific models to derive predictions or decisions based on input data.

The term'machine learning' can be used interchangeably with the term'machine learning'.

As to how to classify data in machine learning, many machine learning algorithms have been developed. Decision trees, Bayesian networks, support vector machines (SVMs), and artificial neural networks (ANNs) are representative.

The decision tree is an analysis method that performs classification and prediction by charting decision rules in a tree structure.

The Bayesian network is a model that expresses a probabilistic relationship (conditional independence) between a number of variables in a graph structure. Bayesian networks are suitable for data mining through unsupervised learning.

The support vector machine is a model of supervised learning for pattern recognition and data analysis, and is mainly used for classification and regression analysis.

An artificial neural network is an information processing system in which a number of neurons, called nodes or processing elements, are connected in the form of a layer structure by modeling the operation principle of biological neurons and the connection relationship between neurons.

Artificial neural networks are models used in machine learning, and are statistical learning algorithms inspired by biological neural networks (especially the brain among animals' central nervous systems) in machine learning and cognitive science.

Specifically, the artificial neural network may refer to an overall model having problem-solving ability by changing the strength of synaptic bonding through learning by artificial neurons (nodes) forming a network by combining synapses.

The term artificial neural network may be used interchangeably with the term neural network.

The artificial neural network may include a plurality of layers, and each of the layers may include a plurality of neurons. In addition, artificial neural networks may include synapses that connect neurons and neurons.

In general, artificial neural networks have three factors: (1) the connection pattern between neurons in different layers, (2) the learning process to update the weight of the connection, and (3) the output value from the weighted sum of the inputs received from the previous layer. It can be defined by the activation function it creates.

The artificial neural network may include network models such as DNN (Deep Neural Network), RNN (Recurrent Neural Network), BRDNN (Bidirectional Recurrent Deep Neural Network), MLP (Multilayer Perceptron), and CNN (Convolutional Neural Network). , Is not limited thereto.

In this specification, the term'layer' may be used interchangeably with the term'layer'.

Artificial neural networks are divided into single-layer neural networks and multi-layer neural networks according to the number of layers.

A typical single-layer neural network consists of an input layer and an output layer.

In addition, a general multilayer neural network is composed of an input layer, one or more hidden layers, and an output layer.

The input layer is a layer that receives external data, the number of neurons in the input layer is the same as the number of input variables, and the hidden layer is located between the input layer and the output layer, receives signals from the input layer, extracts characteristics, and transfers them to the output layer. do. The output layer receives a signal from the hidden layer and outputs an output value based on the received signal. The input signal between neurons is multiplied by each connection strength (weight) and then summed. If this sum is greater than the neuron's threshold, the neuron is activated and the output value obtained through the activation function is output.

Meanwhile, a deep neural network including a plurality of hidden layers between an input layer and an output layer may be a representative artificial neural network that implements deep learning, a type of machine learning technology.

Meanwhile, the term'deep learning' can be used interchangeably with the term'deep learning'.

The artificial neural network can be trained using training data. Here, learning refers to the process of determining the parameters of the artificial neural network using training data in order to achieve the purpose of classification, regression, or clustering of input data. I can. Representative examples of parameters of an artificial neural network include weights applied to synapses or biases applied to neurons.

The artificial neural network learned by the training data may classify or cluster input data according to a pattern of the input data.

Meanwhile, the artificial neural network trained using the training data may be referred to as a trained model in this specification.

The following describes the learning method of artificial neural networks.

Learning methods of artificial neural networks can be broadly classified into supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.

Supervised learning is a method of machine learning to infer a function from training data.

In addition, among the functions to be inferred, outputting a continuous value may be referred to as regression, and predicting and outputting a class of an input vector may be referred to as classification.

In supervised learning, an artificial neural network is trained with a label for training data.

Here, the label may mean a correct answer (or result value) that the artificial neural network must infer when training data is input to the artificial neural network.

In this specification, when training data is input, the correct answer (or result value) to be inferred by the artificial neural network is referred to as a label or labeling data.

In addition, in this specification, setting a label on training data for learning an artificial neural network is referred to as labeling the training data with labeling data.

In this case, the training data and the label corresponding to the training data constitute one training set, and may be input to the artificial neural network in the form of a training set.

Meanwhile, the training data represents a plurality of features, and labeling of the training data may mean that a label is attached to the feature represented by the training data. In this case, the training data may represent the characteristics of the input object in the form of a vector.

The artificial neural network can infer a function for the correlation between the training data and the labeling data by using the training data and the labeling data. In addition, parameters of the artificial neural network may be determined (optimized) through evaluation of a function inferred from the artificial neural network.

Unsupervised learning is a type of machine learning, and there is no label for training data.

Specifically, the unsupervised learning may be a learning method in which an artificial neural network is trained to find and classify patterns in the training data itself, rather than an association relationship between training data and a label corresponding to the training data.

Examples of unsupervised learning include clustering or independent component analysis.

In this specification, the term'clustering' may be used interchangeably with the term'clustering'.

Examples of artificial neural networks using unsupervised learning include Generative Adversarial Network (GAN) and Autoencoder (AE).

A generative adversarial neural network is a machine learning method in which two different artificial intelligences compete and improve performance, a generator and a discriminator.

In this case, the generator is a model that creates new data and can create new data based on the original data.

In addition, the discriminator is a model that recognizes a pattern of data, and may play a role of discriminating whether the input data is original data or new data generated by the generator.

In addition, the generator learns by receiving data that cannot be deceived by the discriminator, and the discriminator can learn by receiving deceived data from the generator. Accordingly, the generator can evolve to deceive the discriminator as well as possible, and the discriminator can evolve to distinguish between the original data and the data generated by the generator.

Auto encoders are neural networks that aim to reproduce the input itself as an output.

The auto encoder includes an input layer, at least one hidden layer and an output layer.

In this case, since the number of nodes in the hidden layer is smaller than the number of nodes in the input layer, the dimension of the data is reduced, and compression or encoding is performed accordingly.

Also, data output from the hidden layer goes to the output layer. In this case, since the number of nodes in the output layer is greater than the number of nodes in the hidden layer, the dimension of the data increases, and accordingly, decompression or decoding is performed.

Meanwhile, the auto-encoder adjusts the connection strength of neurons through learning, so that input data is expressed as hidden layer data. In the hidden layer, information is expressed with fewer neurons than the input layer, but the fact that the input data can be reproduced as an output means that the hidden layer found and expressed a hidden pattern from the input data.

Semi-supervised learning is a kind of machine learning, and may mean a learning method that uses both labeled training data and unlabeled training data.

As one of the techniques of semi-supervised learning, there is a technique of inferring a label of training data that is not given a label and then performing learning using the inferred label. This technique is useful when the cost for labeling is high. I can.

Reinforcement learning is the theory that given an environment in which an agent can judge what action to do at every moment, it can find the best way to experience without data.

Reinforcement learning can be performed mainly by the Markov Decision Process (MDP).

Explaining the Markov decision process, first, an environment is given where the information necessary for the agent to perform the next action is given, secondly, it defines how the agent will behave in that environment, and thirdly, what the agent does well is rewarded ( Reward) is given and the penalty is given for failing to do something, and fourthly, the optimal policy is derived through repeated experiences until the future reward reaches the highest point.

The structure of the artificial neural network is specified by the configuration of the model, activation function, loss function or cost function, learning algorithm, optimization algorithm, etc., and hyperparameters are pre-trained. It is set, and then, a model parameter is set through learning, so that the content can be specified.

For example, factors that determine the structure of an artificial neural network may include the number of hidden layers, the number of hidden nodes included in each hidden layer, an input feature vector, a target feature vector, and the like.

Hyperparameters include several parameters that must be initially set for learning, such as initial values of model parameters. And, the model parameter includes several parameters to be determined through learning.

For example, the hyperparameter may include an initial weight value between nodes, an initial bias value between nodes, a mini-batch size, a number of learning iterations, a learning rate, and the like. In addition, the model parameter may include a weight between nodes, a bias between nodes, and the like.

The loss function can be used as an index (reference) for determining an optimal model parameter in the learning process of the artificial neural network. In artificial neural networks, learning refers to the process of manipulating model parameters to reduce the loss function, and the purpose of learning can be seen as determining model parameters that minimize the loss function.

The loss function may mainly use a mean squared error (MSE) or a cross entropy error (CEE), but the present invention is not limited thereto.

The cross entropy error may be used when the correct answer label is one-hot encoded. One-hot encoding is an encoding method in which the correct answer label value is set to 1 only for neurons corresponding to the correct answer, and the correct answer label value is set to 0 for non-correct answer neurons.

In machine learning or deep learning, learning optimization algorithms can be used to minimize loss functions, and learning optimization algorithms include gradient descent (GD), stochastic gradient descent (SGD), and momentum. ), NAG (Nesterov Accelerate Gradient), Adagrad, AdaDelta, RMSProp, Adam, Nadam, etc.

Gradient descent is a technique that adjusts model parameters in the direction of reducing the loss function value by considering the slope of the loss function in the current state.

The direction to adjust the model parameter is referred to as the step direction, and the size to be adjusted is referred to as the step size.

In this case, the step size may mean a learning rate.

In the gradient descent method, a gradient is obtained by partially differentiating a loss function into each model parameter, and the model parameters may be updated by changing the acquired gradient direction by a learning rate.

Probabilistic gradient descent is a technique that increases the frequency of gradient descent by dividing training data into mini-batch and performing gradient descent for each mini-batch.

Adagrad, AdaDelta, and RMSProp are techniques that increase optimization accuracy by adjusting the step size in SGD. In SGD, momentum and NAG are techniques to increase optimization accuracy by adjusting the step direction. Adam is a technique that improves optimization accuracy by adjusting the step size and step direction by combining momentum and RMSProp. Nadam is a technique that improves optimization accuracy by adjusting step size and step direction by combining NAG and RMSProp.

The learning speed and accuracy of an artificial neural network are highly dependent on hyperparameters, as well as the structure of the artificial neural network and the type of learning optimization algorithm. Therefore, in order to obtain a good learning model, it is important not only to determine an appropriate artificial neural network structure and learning algorithm, but also to set appropriate hyperparameters.

In general, hyperparameters are experimentally set to various values to train an artificial neural network, and as a result of the learning, the hyperparameter is set to an optimal value that provides a stable learning speed and accuracy.

2 is a flowchart illustrating a method of operating a refrigerator according to an exemplary embodiment.

The control unit 150 may receive operation data related to the operation state of the refrigerator. The control unit 150 may receive operation data including a factor indicating the state of each component of the refrigerator from various sensors provided in the refrigerator (S110).

As described above, the factor may be a current applied to the compressor 110, a temperature at each component of the refrigerator, a pressure ratio of the compressor 110, an opening degree of the hot gas valve 181, and the like.

In the step of determining whether a surge has occurred, the controller 150 may determine whether the compressor 110 provided in the refrigerator is in a normal state, an impending surge state, or a surge occurrence state from the operation data ( S120).

Meanwhile, the impending surge occurrence state may be defined as, for example, a state from a time point before a set time from a surge occurrence point of the compressor 110 to the surge occurrence point. For example, if the current is in a normal state, but a surge occurs 10 minutes later, the time from 10 minutes before the occurrence of the surge to the time of occurrence of the surge may be defined as the impending surge occurrence state.

However, the set time is not limited to 10 minutes, and may be appropriately selected in consideration of the characteristics of the refrigerator and a control method for preparing for the occurrence of surge.

In the artificial intelligence model learning to be described later, the processor can grasp characteristic values of factors in each of the surge occurrence state and the impending surge occurrence state, taking into account this time difference between the surge occurrence state and the impending surge occurrence state. Accordingly, the processor may derive comparison data that distinguishes whether the compressor is in a surge generation state or an impending surge generation state.

In the embodiment, whether the compressor 110 is in a steady state, an impending surge occurrence state, or a surge occurrence state, when the operation data is compared with the comparison data, the factors of the operation data are the normal region of the comparison data, the surge occurrence impending region, or It means that it is located in the surge generation area.

In determining whether a surge has occurred, the controller 150 may determine whether at least one factor included in the operation data is located in a normal region of the comparison data, an impending surge region, or a surge occurrence region.

The comparative data can be retained by dividing the values of various factors that can be obtained from the refrigerator into a normal region, an impending surge region, or a surge occurrence region. The comparison data may be derived by learning according to an artificial intelligence model using a support vector machine.

The artificial intelligence model of the embodiment uses a support vector machine, and the support vector machine may set a boundary between a normal region of the comparison data, an impending surge region, or a surge occurrence region.

The support vector machine recognizes the factor as a vector, and may set a boundary to distinguish whether each factor is located in a normal region, an impending surge region, or a surge occurrence region by learning.

Learning the artificial intelligence model may be performed in the processor 160. The processor 160 may be provided in the server. The server may receive the driving data from the refrigerator, and the processor 160 may derive the comparison data by performing artificial intelligence model learning based on the received driving data.

The comparison data is transmitted to the refrigerator again, and the control unit 150 may determine the surge of the compressor 110 by comparing the comparison data with the operation data of the refrigerator currently in operation and the comparison data.

Further, the processor 160 may perform learning according to an artificial intelligence model using the driving data, but may derive the comparison data corresponding to each cooling capacity of the refrigerator.

The operation mode of the refrigerator may be different according to the refrigeration capacity, or different models of refrigerators may be used depending on the refrigeration capacity. Accordingly, in the embodiment, all different comparison data are derived according to the refrigeration capacity of the refrigerator, so that even when the operation mode of the refrigerator is different or the model of the refrigerator is different, the comparison data derived by artificial intelligence model learning can be used.

Hereinafter, learning of an artificial intelligence model using a support vector machine will be described in detail.

3 is a flow chart illustrating learning an artificial intelligence model according to an embodiment.

The operator may select factors that serve as a criterion for determining whether a surge is generated from the operation data (S210). Various factors are included in the operation data, and among these factors, it is necessary to select factors sensitive to the surge in the compressor 110.

For example, as the numerical value changes, factors in which the compressor 110 surges are frequently generated are the target of learning, and even if the numerical value is changed, the factors having almost no surge generation of the compressor 110 are excluded from the target of learning, artificial intelligence. It can reduce the time of model training and increase the efficiency.

Data necessary for factor selection may be included in the operation data. The data required for factor selection may include values of various factors and whether a surge in the compressor 110 is generated.

The operator may select factors in which surges are frequently generated in the compressor 110 as the numerical value changes from the data required for factor selection.

In the selection of factors, for example, algorithms such as Principle Component Analysis (PCA), Minimum Redundancy Maximum Relevance (mRMR), Wrapper Methods, Chi-Squared Test, and F1 score may be used.

In the selection of factors, the operator determines the order of the factors in the order in which the frequency of surge generation is high as the numerical value changes for a plurality of factors included in the operation data, and the number is set sequentially from the highest priority factors. Can be selected.

For example, the operator can quantify the frequency of surge occurrence due to a change in numerical value for a plurality of factors as an F1 score. The F1 score is a data evaluation index and is expressed as a harmonic mean of precision and recall. The F1 score is a known data evaluation index and is obvious to a person skilled in the art, so a detailed description thereof will be omitted.

In the following table, using some of the driving factors included in the driving data, the frequency of surge occurrence due to the change in the numerical value of a plurality of factors is numerically converted into an F1 score, and the frequencies are listed in the order of high frequency.

Factor rank factor F1 score One Current applied to the compressor 241.94 2 Compressor inlet refrigerant temperature 181.57 3 Printed circuit board temperature of compressor mother rate 2.50 4 The temperature of the motor bearing of the compressor 1.93 5 Evaporator pressure 1.81 6 Frequency of alternating current applied to the motor of the compressor 1.12 7 Compressor pressure ratio 0.11 8 Compressor outlet refrigerant temperature 0.04 9 Hot gas valve opening 0.02 10 Condenser pressure 0.01

For example, if two factors to be used for artificial intelligence model training are selected based on the F1 score, it is necessary to select the current applied to the compressor 110 and the inlet refrigerant temperature of the compressor 110 with the highest ranking on the F1 score. May be appropriate.

Accordingly, the factor used for learning the artificial intelligence model may include, for example, at least one of a current applied to the compressor 110 or a refrigerant temperature at an inlet of the compressor 110.

However, the present invention is not limited thereto, and the algorithm used for factor selection and the number and selection of factors used for learning the artificial intelligence model may be variously selected.

Labeling data may be derived from the selected factors (S220). The labeling data is data in which the correct answer (or result value) to be inferred by the support vector machine is designated. Since AI model learning using a support vector machine is supervised learning, labeling data is required.

The labeling data may be in a state in which a boundary between a normal region, an imminent surge occurrence region, or a surge occurrence region is set for a plurality of factors.

Derivation of labeling data may be performed, for example, as a separate process. A task for deriving labeling data, that is, labeling, may be performed by an operator or may be performed by a separate machine learning.

The derivation of the labeling data may be performed in a manner that specifically selects a part of the driving data and classifies the selected driving data into one of a normal region, an imminent surge occurrence region, or a surge occurrence region.

Labeling data prepared by such a labeling operation may be input to the processor 160 for learning an artificial intelligence model.

Regardless of whether it is carried out by an operator or machine learning, it is appropriate to save time as much as possible for labeling and proceed in the direction of deriving labeling data with clear boundaries between areas.

The operator may proceed with the data pre-processing step (S230). The data pre-processing step may include a process of normalizing training data provided with factors selected from the driving data. Training data means data on driving factors selected from driving data.

Since the training data is data measured from various sensors, erroneous data and unnecessary data may be included. Therefore, it is necessary to remove these erroneous data and unnecessary data, which is possible by normalizing the training data.

In addition, since each factor of the training data has different units and sizes, it is necessary to convert the numerical values of the training data into a certain range. To this end, in the data preprocessing step, values of each factor of the training data may be converted into values between 0 and 1 through, for example, Min-Max Normalization.

Since the above-described minimum-maximum normalization itself is apparent to those of ordinary skill in the art, detailed descriptions of the normalization process are omitted herein.

The data preprocessing step may include deleting missing values from the training data. The missing value is a numerical value of 0, and if the training data contains the missing value, it may take a long time to train the artificial intelligence model, and the accuracy of the learning result may be lowered.

Accordingly, in an embodiment, by removing missing values from training data used for learning an artificial intelligence model, it is possible to reduce the learning time and increase the accuracy of the learning result.

The operator can optimize the artificial intelligence model using the support vector machine (S240). In step S240, a hyperparameter may be selected.

The hyperparameter may be an initial value of factors set for learning an artificial intelligence model. Depending on the selection of hyperparameters, the time it takes to train the artificial intelligence model and the results of the learning may vary.

Therefore, the operator needs to select an appropriate hyperparameter in order to shorten the time it takes to learn the artificial intelligence model as much as possible and to derive comparative data with high accuracy. For example, an operator can select a hyperparameter by referring to the labeling data.

Learning the artificial intelligence model may be performed by the processor 160. The processor 160 may perform training according to the artificial intelligence model optimized using the training data (S250).

In step S250, the labeling data and the training data may be input to the processor 160 for learning the artificial intelligence model as one data set. Learning can be performed by an artificial intelligence learning model using a support vector machine.

The processor 160 may perform learning and set a boundary of a normal region, an imminent surge occurrence region, or a surge occurrence region in the training data based on the labeling data.

For example, the processor 160 may recognize a value of a factor included in each region as a vector, and may set a boundary for each region to maximize a margin. In this case, the margin may mean the shortest distance among the distances from the boundary dividing the two regions to factors belonging to each region.

In an embodiment, the area where the factors are located is divided into a total of 3, and the processor 160 may perform learning by binarizing each area. Accordingly, the processor 160 may set the boundary of any one of the normal region, the surge generation imminent region, or the surge generation region in the training data, and then sequentially set the boundary of the remaining two regions.

4 is a graph showing the distribution of factors related to surges of the compressor 110 in learning an artificial intelligence model according to an embodiment. In FIG. 4, for example, a graph is shown using values of two factors, but a boundary may be defined for one or three or more factors.

With respect to the distribution of factors illustrated in FIG. 4, the processor 160 may perform learning to select a boundary for dividing each region. Since each region is binary-coded to perform learning, for example, the processor 160 may sequentially set a boundary for dividing a surge occurrence region, a boundary for dividing an impending surge region, and a boundary for dividing a normal region.

FIG. 5 is a diagram illustrating a boundary for dividing a surge generation region and a remaining region in the graph of FIG. 4.

6 is a diagram illustrating a boundary for dividing an area that is about to generate a surge and the remaining areas in the graph of FIG.

7 is a diagram illustrating a boundary dividing a normal region from a remaining region in the graph of FIG. 4.

As shown in FIG. 5, the processor 160 may perform learning to set a boundary for dividing the surge generation region and the remaining regions for each factor.

Next, as shown in FIG. 6, the processor 160 may perform learning to set a boundary for dividing the region of imminent surge occurrence and the remaining region for each factor.

Next, as shown in FIG. 7, the processor 160 may perform learning to set a boundary for dividing the normal region and the remaining region for each factor.

Through the above process, the processor 160 may set a boundary for dividing an area in which factors of the training data are located into a normal area, an impending surge area, and a surge generation area.

The processor 160 may perform learning with the input training data and derive comparison data as a result of the learning. The comparison data may be derived from the training data in a state in which a boundary of a normal region, an imminent surge occurrence region, or a surge occurrence region is set.

The comparison data is transmitted to the control unit 150, and the control unit 150 may determine whether a surge is generated by the compressor 110 by comparing the comparison data with the currently input operation data.

Referring back to FIG. 1, the refrigerator may further include a memory 170 for storing comparison data. The controller 150 may determine whether a surge is generated by the compressor 110 based on the comparison data stored in the memory 170.

In an embodiment, the processor 160 may perform artificial intelligence model training at any time when the refrigerator is operated, and comparison data changed according to the learning result may be updated in the memory 170.

8 is a flow chart showing the determination of the surge of the compressor 110 in the refrigerator according to an embodiment.

2 and 8, the control unit 150 may determine the surge of the compressor 110 by comparing the input current operation data and the comparison data stored in the memory 170 with each other. That is, the control unit 150 may determine in which region of the comparison data at least one factor included in the driving data is located in a normal region, an imminent surge occurrence region, or a surge occurrence region.

First, the control unit 150 may determine whether the driving data is in the surge generation region. If the operation data is in the surge generation region, the control unit 150 may determine that the compressor 110 has a surge occurrence.

If the driving data is not in the surge generation area, then the controller 150 may determine whether the driving data is in the surge generation imminent area. If the operation data is in a surge generation imminent region, the controller 150 may determine that the surge generation is imminent with respect to the compressor 110.

If the operation data is not in the region where the surge generation is imminent, the controller 150 may determine that the compressor 110 is operating normally.

The determination of whether or not a surge is generated may be performed at a set number of times within a set time. For example, the control unit 150 may determine whether or not a surge is generated 10 times for 10 seconds.

The control unit 150 may select a state with the highest frequency among the 10 determinations. For example, if it is determined that the normal state is 2, the surge generation state is 3, and the surge generation impending state is 5 among the 10 determinations, the controller 150 may determine that the surge generation is imminent.

If the number of times of each state is the same, the control unit 150 may determine a state capable of blocking the occurrence of a surge without stopping the operation of the refrigerator. If there is a surge occurrence, the operation may need to be stopped, so the impending surge condition can be prioritized. In addition, since it is advantageous to block the occurrence of a surge, a surge occurrence impending state or a surge occurrence state can be prioritized over a normal state.

Referring back to FIG. 2, the controller 150 may perform step S110 again when the compressor 110 is in a surge impending state or is not in a surge generating state.

When the compressor 110 is in a surge occurrence state or a surge occurrence state, the controller 150 may control to block the surge generation (S130).

When a surge is imminent or a surge is generated, for example, the controller 150 opens the hot gas valve 181 and increases the opening degree to increase the flow rate of the refrigerant flowing into the compressor 110, or the compressor ( 110) to reduce the pressure ratio of the compressor 110, it is possible to block the generation of the surge.

Meanwhile, in the case of a surge occurrence state, the controller 150 may stop the operation of the compressor 110 in order to prevent damage to the compressor 110 due to the surge.

When the compressor 110 is in a surge occurrence state or a surge occurrence state, the control unit 150 provided in the refrigerator may output notification information (S140).

The notification information may be output through a user interface provided in the refrigerator or various devices connected to communicate with the refrigerator, for example, a user's computer, a mobile device, and the like and transmitted to the user. In this case, the control unit 150 may divide and output a surge generation impending state and a surge generation state.

The user can receive notification information visually or aurally, such as blinking of text, sound, and blinking. Users can receive notification information from the freezer and take quick and effective measures for the impending surge or surge occurrence condition.

In the embodiment, since it is possible to determine the occurrence of the surge in the compressor 110 by learning the artificial intelligence model, it is possible to significantly reduce the occurrence of a judgment error compared to determining the occurrence of a surge by another device.

In the embodiment, since it is possible to determine about the occurrence of a surge in the compressor 110 using only information obtained from various sensors already provided in the refrigerator without using a separate device, there is an effect of reducing the production cost of the refrigerator. .

As described above with respect to the embodiments, only a few are described, but other various forms of implementation are possible. The technical contents of the above-described embodiments may be combined in various forms unless they are technologies incompatible with each other, and may be implemented as a new embodiment through this.

110: compressor
120: condenser
130: expansion device
140: evaporator
150: control unit
160: processor
170: memory
180: hot gas line
181: hot gas valve

Claims (13)

Receiving operation data related to an operation state of the refrigerator;
Determining whether a surge is generated from the operation data to determine whether a compressor provided in the refrigerator is in a normal state, an impending surge state, or a surge occurrence state;
When the compressor is in a surge generation impending state or a surge generation state, controlling to block the generation of the surge; And
Outputting notification information when the compressor is in a surge occurrence state or a surge occurrence state
Including,
The judgment of whether a surge has occurred is
It is determined whether at least one factor included in the operation data is located in a normal region, an imminent surge occurrence region, or a surge occurrence region of the comparison data,
The comparison data,
It is derived by learning according to an artificial intelligence model using a support vector machine,
The support vector machine,
Setting a boundary between a normal region, an imminent surge occurrence region, or a surge occurrence region of the comparison data,
How to operate the freezer. The method of claim 1,
The freezer,
The compressor;
A condenser whose inlet is connected to the outlet of the compressor;
An expansion device in which an inlet is connected to an outlet of the condenser;
An evaporator having an inlet connected to an outlet of the expansion device and an outlet connected to an inlet of the compressor; And
A control unit that controls the operation of the refrigerator
Containing, the refrigerator operating method. The method of claim 2,
The control unit is connected to a processor for deriving the comparison data,
The processor,
A method of operating a refrigerator, wherein the learning is performed according to an artificial intelligence model using the operation data, and the comparison data corresponding to each of the cooling capacity of the refrigerator is derived. The method of claim 1,
Artificial intelligence model learning,
Selecting factors as a criterion for determining whether a surge has occurred in the operation data;
Deriving labeling data from the selected factors;
A data pre-processing step including normalizing training data provided with factors selected from the driving data;
Optimizing an artificial intelligence model using a support vector machine;
Performing training according to the artificial intelligence model optimized using the training data
Containing, the refrigerator operating method. The method of claim 4,
The choice of factors is,
For a plurality of factors included in the operation data, as the numerical value changes, the order of the factors is determined in the order in which the frequency of surge occurrence is high, and the number of factors that are set sequentially from the highest priority factors is selected. The method of claim 4,
The derivation of labeling data is,
Selecting a part of the operation data, and classifying the selected operation data into any one of a normal region, an impending surge region, or a surge generation region. The method of claim 6,
The performance of learning is,
Based on the labeling data, setting a boundary between a normal region, an impending surge region, or a surge occurrence region in the training data. The method of claim 7,
The performance of learning is,
In the training data, after setting a boundary of any one of a normal region, an impending surge occurrence region, or a surge occurrence region, the boundary of the remaining two regions is sequentially set. The method of claim 8,
The comparison data is derived from the training data in a state in which a boundary of a normal region, an imminent surge occurrence region, or a surge occurrence region is set,
A method of operating a refrigerator, wherein the determination of whether or not a surge is generated is performed at a set number of times within a set time. The method of claim 4,
Data preprocessing is,
And deleting missing values from the training data. The method of claim 4,
Optimization of the artificial intelligence model,
A method of operating a refrigerator, comprising selecting a hyperparameter. The method of claim 1,
The state of imminent surge occurrence,
A method of operating a refrigerator, which is defined as a state from a time point before a set time from a time when a surge occurs of the compressor to a time point at which the surge occurs. The method of claim 1,
The factor is,
A method of operating a refrigerator comprising at least one of a current applied to the compressor or a temperature of a refrigerant at an inlet of the compressor.

Images