Air conditioner and control method

Abstract

The invention relates to the technical field of air conditioners, in particular to an air conditioner which comprises a machine shell, a compressor, a draught fan and a control unit, and the control unit is configured to judge whether the air conditioner is in a starting-up and compressor stopping state or not in response to a first preset duration; if the starting state and the compressor stopping state are met, the air conditioner is controlled to operate in the preset air supply mode for a second preset duration, and actual operation data are obtained; on the basis of the actual operation data and a predetermined standard operation data set, standard operation data corresponding to the closest standard working condition is obtained in a matched mode and serves as reference operation data, and the reference operation data comprises multi-level filth blockage power differences; and inputting the actual operation data and the reference operation data into a double-path discrimination network model, performing feature extraction and feature fusion respectively, performing feature recognition through the double-path discrimination network model, outputting probability distribution of each filth blockage level according to the multi-level filth blockage power difference, and obtaining the filth blockage level of the air conditioner.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning, and more particularly, to an air conditioner and a control method thereof. Background Art

[0002] Currently, the blockage detection function of current air conditioning systems is highly dependent on the static pressure adaptive module. Static pressure adaptive technology determines the degree of filter or heat exchanger blockage by monitoring changes in air duct pressure, but it has significant limitations. Some models without static pressure detection modules are unable to implement blockage detection. Even models equipped with static pressure adaptive function still rely on the function selection trigger of preset cycles for blockage detection. User permissions are closed, parameter configuration is complex, the operation threshold is high, and it only supports fixed-cycle detection. It lacks real-time dynamic monitoring capabilities and cannot respond to sudden blockages. Summary of the Invention

[0003] The present application provides an air conditioner and a control method to at least solve the problems in the related art, such as the inability of non-static pressure models to identify dirt and blockage due to reliance on a static pressure adaptive module, the lack of real-time dynamic monitoring capabilities due to the existing detection mechanism being limited to a fixed trigger cycle, and the high operating threshold caused by closed user permissions and complex parameter configuration.

[0004] In a first aspect, the present application provides an air conditioner, comprising:

[0005] The casing has an air outlet and a return air inlet provided on its periphery;

[0006] A compressor and a fan are arranged in the casing;

[0007] a control unit electrically connected to the compressor and the fan;

[0008] The control unit is configured to:

[0009] In response to a first preset time, determining whether the air conditioner is in an on-state and the compressor is in a stopped state;

[0010] If the power-on state is met and the compressor is stopped, controlling the air conditioner to operate in a preset air supply mode for a second preset time period, and obtaining actual operating data;

[0011] Based on the actual operating data and a predetermined standard operating data set, matching and obtaining the standard operating data corresponding to the closest standard operating condition as the benchmark operating data, wherein the benchmark operating data includes multi-level dirty and blocked power differences;

[0012] The actual operation data and the benchmark operation data are input into a dual-path discriminant network model, feature extraction is respectively performed after feature fusion, and feature recognition is performed through the dual-path discriminant network model, a probability distribution of each dirty blockage level is output according to the multi-grade dirty blockage power difference, and the dirty blockage level of the air conditioner is obtained.

[0013] In the technical scheme, dirty blockage recognition is realized by replacing static pressure detection with fan motor power variation, the problem of missing dirty blockage monitoring of a static pressure-free model is solved, a fixed cycle detection lag defect is overcome by capturing a compressor shutdown interval to trigger a real-time detection mechanism, real-time response to sudden dirty blockage is realized, a lightweight dual-path discriminant network model is used to automatically complete working condition matching, feature fusion and level determination in a closed loop process without user intervention, operation complexity and operation and maintenance cost are significantly reduced, and equipment performance stability is ensured.

[0014] In some embodiments of the present application, the control unit is further configured to:

[0015] Obtain operation history data of the air conditioner in a non-dirty blockage state, the operation history data including a standard indoor temperature ti, a standard indoor humidity hi and a standard fan power;

[0016] Generate n groups of standard working conditions based on clustering of the operation history data, simulate 1-m grade dirty blockage states under a preset air supply mode for each group of standard working conditions and obtain dirty blockage fan powers Pm (m = 1, 2,..., m), and calculate multi-grade dirty blockage power differences {P0, △P1, △P2,..., △Pm} based on a standard fan power P0 and the dirty blockage fan powers Pi.

[0017] Construct a standard operation data set based on the operation history data and the multi-grade dirty blockage power differences, and the standard operation data set is denoted as {ti, hi, P0, △P1, △P2,..., △Pm}.

[0018] In the technical scheme, a standard data set covering various actual working conditions is automatically constructed by clustering user historical operation data, the universality and accuracy of dirty blockage determination are significantly improved, a power difference quantification mechanism for simulating multi-grade dirty blockage states under a preset air supply mode is combined to establish a dirty blockage level mapping benchmark, and dynamic difference calculation of a standard fan power and a dirty blockage power eliminates environmental variable interference and solves the defects of high misjudgment rate and poor model adaptability caused by traditional manual setting of a static threshold.

[0019] In some embodiments, the control unit is further configured to:

[0020] The actual operation data includes a first indoor temperature, a first indoor humidity and a first fan power;

[0021] The actual operation data is matched with the standard operation data set according to the first indoor temperature and the first indoor humidity based on a weighted distance algorithm, and standard operation data corresponding to the closest standard working condition is obtained as reference operation data, the reference operation data including a second indoor temperature, a second indoor humidity, a second fan power and a multi-grade dirty block power difference.

[0022] In the technical scheme, the real-time temperature and humidity are dynamically matched with the standard working condition library through the weighted distance algorithm to lock the optimal reference operation data, so that the problem of dirty block misjudgment caused by environmental fluctuations is solved; the model generalization capability and the dirty block identification accuracy in a complex use scenario are improved by taking the temperature and humidity as the judgment reference.

[0023] In some embodiments, the control unit is further configured to:

[0024] During initial installation, the fan is controlled to operate in a preset air supply mode to stabilize under a standard working condition, and a real-time power value is recorded;

[0025] A compensation value is generated according to a difference between the standard fan power corresponding to the standard working condition and the real-time power value, and all standard fan powers in the standard operation data set are corrected.

[0026] In the technical scheme, the motor power reference value is calibrated in real time during the initial installation stage to eliminate errors caused by individual differences of the motor; the standard data set is corrected based on the compensation value, and a dirty block judgment reference conforming to the actual air conditioner is constructed, so that the identification results of different devices under the same working condition remain high consistency, and the dirty block identification accuracy is improved.

[0027] In some embodiments, the control unit is further configured to:

[0028] The dual-path judgment network model includes an actual operation data feature extraction branch and a reference operation data feature extraction branch;

[0029] The actual operation data is input into an input layer of the actual operation data feature extraction branch, and is sequentially processed through a full connection layer, a ReLU activation function layer and a Dropout layer to obtain an actual feature vector;

[0030] The reference operation data is input into an input layer of the reference operation data feature extraction branch, and is sequentially processed through a full connection layer, a ReLU activation function layer and a Dropout layer to obtain a reference feature vector.

[0031] In the technical scheme, the actual operation state and the benchmark working condition are decoupled in depth by a double-branch feature extraction architecture, so that the recognition granularity of the dirty blockage feature is improved; a nonlinear mapping space is constructed by using a full connection layer and a ReLU activation function to extract a dirty blockage sensitive factor in a high-dimensional working condition feature; and a generalization capability is enhanced by a Dropout layer.

[0032] In some embodiments, the control unit is further configured to:

[0033] The actual operation data and the benchmark operation data are normalized respectively, and then input into the actual operation data feature extraction branch and the benchmark operation data feature extraction branch respectively.

[0034] In the technical scheme, the actual operation data and the benchmark operation data are normalized, so that the value magnitude difference of multiple source parameters such as temperature, humidity and power is eliminated, and standardized input is provided for the double-path discriminant network.

[0035] In some embodiments, the double-path discriminant network model further comprises a double-path feature fusion branch.

[0036] The actual feature vector and the benchmark feature vector are input into the double-path feature fusion branch for feature splicing, and then sequentially pass through a first full connection layer, a ReLU activation function layer, a second full connection layer and a Softmax activation function, to output a probability distribution of the dirty blockage level of the air conditioner, so as to obtain the dirty blockage level.

[0037] In the technical scheme, the deep features of the actual working condition and the benchmark working condition are fused by feature splicing, so that the perception of the model to the dirty blockage difference is improved; a high-robustness dirty blockage decision boundary is constructed by cascading processing of the double full connection layers and the nonlinear activation, so as to capture the dynamic coupling relationship of the power difference and the environmental parameters; and finally, the dirty blockage level is output by the Softmax probability distribution, so as to realize the recognition and level determination of the dirty blockage state.

[0038] In some embodiments, the control unit is further configured to: if the dirty blockage level reaches a preset level, the control unit controls the air conditioner to display an alarm.

[0039] In the technical scheme, the intelligent alarm mechanism is triggered by the preset dirty blockage level threshold, so as to realize real-time alarm of the moderate and severe dirty blockage that needs to be intervened, so that the user can respond to the dirty blockage in time, and the energy consumption surge and equipment damage caused by the dirty blockage problem are effectively avoided.

[0040] In some embodiments, the control unit is further configured to:

[0041] The control unit controls the air conditioner to restore the preset operation mode after obtaining the actual operation data.

[0042] In the technical solution, the compressor stop gap triggering real-time detection mechanism is captured, and after the actual operation data is obtained, the user preset operation mode is switched back, the interference of traditional dirty block detection on normal operation of the air conditioner is eliminated, room temperature fluctuation caused by mode mutation is avoided, user experience is ensured, and energy saving effect optimization is realized.

[0043] In a second aspect, the application provides an air conditioner control method, comprising:

[0044] In response to a first preset time length, it is judged whether the air conditioner is in a state of starting and compressor stopping;

[0045] If the state of starting and compressor stopping is met, the air conditioner is controlled to run in a preset air supply mode for a second preset time length, and actual operation data is obtained;

[0046] Based on the actual operation data and a standard operation data set determined in advance, the closest standard operation data corresponding to a standard working condition is matched and obtained as reference operation data, wherein the reference operation data includes multi-grade dirty block power difference;

[0047] The actual operation data and the reference operation data are input into a double-path discriminant network model, feature extraction is performed respectively after feature fusion, and feature recognition is performed through the double-path discriminant network model, the probability distribution of each dirty block grade is output according to the multi-grade dirty block power difference, and the dirty block grade of the air conditioner is obtained. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 is a schematic diagram of an air conditioner structure according to the embodiment of the application;

[0049] Figure 2 is a running flowchart of a control unit according to the embodiment of the application;

[0050] Figure 3 is a flowchart of constructing a standard operation data set according to the embodiment of the application;

[0051] Figure 4 is a standard fan power correction flowchart according to the embodiment of the application;

[0052] Figure 5 is a schematic diagram of actual operation data feature extraction branch structure according to the embodiment of the application;

[0053] Figure 6 is a schematic diagram of reference operation data feature extraction branch structure according to the embodiment of the application;

[0054] Figure 7 is a schematic diagram of double-path feature fusion branch structure according to the embodiment of the application;

[0055] Figure 8 is a schematic diagram of a dual-path discrimination network model structure according to an embodiment of the present application;

[0056] Figure 9 is a module linkage flowchart according to an embodiment of the present application;

[0057] Figure 10 is a hardware configuration diagram of a control unit according to an embodiment of the present application.

[0058] In the above figures:

[0059] 1, housing; 2, compressor; 3, fan; 4, ambient temperature sensor; 5, ambient humidity sensor; 6, control unit; 80, bus; 81, processor; 82, memory; 83, communication interface. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0061] In the description of the present application, it should be understood that the terms "center", "upper", "lower", "front", "back", "left", "right", "vertical", "liquid level", "top", "bottom", "inner", "outer", and the like indicate the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. The terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features limited by "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0062] In the description of the present application, it should be noted that, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0063] In a multi-split air conditioning system, the identification of the dirty and blocked state of the indoor heat exchanger and air duct is a key technology to ensure efficient operation of the equipment.

[0064] The current mainstream solution highly depends on a static pressure adaptive module to realize dirty and blocked detection. However, for four-direction air outlet, one-direction air outlet, and wall-mounted machines without static pressure adaptive function, due to the lack of static pressure sensors and corresponding algorithm support, the dirty and blocked conditions such as heat exchanger dust and air duct blockage cannot be automatically identified, which leads to problems such as refrigeration and heating efficiency decay, energy consumption increase, and fan load abnormality when the equipment is running in a dirty and blocked condition for a long time.

[0065] Even for some models with static pressure adaptive function, the dirty and blocked detection mechanism still has significant limitations. The static pressure adaptive function usually needs to trigger the detection program through a pre-set time function selection module, and the function selection module usually involves user permission hierarchical management, and the parameter configuration needs to be completed by professional personnel through a specific tool, which has high operation complexity and insufficient openness.

[0066] At the same time, the dirty and blocked identification scheme based on the static pressure adaptive function has a fixed detection period and cannot respond to sudden dirty and blocked conditions in real time, which may lead to continuous deterioration of the dirty and blocked problem within the fixed period, affecting the reliability of the equipment.

[0067] Therefore, there is an urgent need for a dirty and blocked identification scheme that does not depend on the static pressure adaptive module and has real-time detection capability.

[0068] Figure 1 As shown in the schematic structure of the air conditioner according to the embodiments of the present application, Figure 1 the air conditioner includes a casing 1.

[0069] In some embodiments, the casing 1 is provided with an air outlet and a return air inlet around the periphery.

[0070] The air outlet adopts a multi-section air deflector structure, and each section of the air deflector can be independently adjusted in angle to realize multi-angle wide-area air supply.

[0071] In some embodiments, the air conditioner includes a compressor 2, which is used to compress low-temperature and low-pressure refrigerant gas into high-temperature and high-pressure gas, to drive the circulation of the refrigerant in the air conditioning system, and to realize the refrigeration or heating function.

[0072] In some embodiments, the air conditioner includes a fan 3, which is used to drive indoor air to enter the inside of the casing 1 through the return air inlet, to exchange heat with the heat exchanger, and to be discharged through the multi-section air deflector structure of the air outlet, to realize air circulation.

[0073] In some embodiments, the air conditioner includes an ambient temperature sensor 4, which is arranged at the return air inlet and is used to collect indoor temperature in real time.

[0074] In some embodiments, the air conditioner comprises an ambient humidity sensor 5 arranged at the return air inlet for real-time acquisition of indoor humidity.

[0075] In some embodiments, the air conditioner can comprise a control unit 6. The control unit 6 is electrically connected to the compressor 2, the fan 3, the ambient temperature sensor 4 and the ambient humidity sensor 5. The control unit 6 obtains the indoor temperature detected by the ambient temperature sensor 4 and the indoor humidity detected by the ambient humidity sensor 5 in real time, synchronously receives the motor power signal of the fan 3 and the start-stop state signal of the compressor 2, and controls the fan 3 to operate in a preset mode.

[0076] In some embodiments, as shown in Figure 2 The control unit 6 judges whether the air conditioner is in the state of starting and compressor stopping in response to a first preset time length.

[0077] If the state of starting and compressor stopping is met, the air conditioner is controlled to operate in a preset air supply mode for a second preset time length to obtain actual operation data.

[0078] Based on the actual operation data and a pre-determined standard operation data set, the standard operation data corresponding to the closest standard working condition is matched and obtained as reference operation data, wherein the reference operation data comprises multi-grade dirty and blocked power difference.

[0079] The actual operation data and the reference operation data are input into a dual-path discriminant network model, feature extraction is performed respectively after feature fusion, and feature recognition is performed through the dual-path discriminant network model, and the dirty and blocked grade of the air conditioner is output.

[0080] Specifically, the control unit 6 performs dirty and blocked detection execution condition judgment in response to a first preset time length, which can be set according to the air conditioner model, the air conditioner operating state and the dirty and blocked recognition requirement.

[0081] Specifically, before performing dirty and blocked detection, the control unit 6 needs to detect that the unit meets two conditions at the same time, i.e., the unit is in the state of starting and the unit is in the state of entering TH off.

[0082] The unit in the state of starting means that the power of the air conditioner has been turned on and the fan 3 keeps running, ensuring that dirty and blocked determination can be performed through the motor power value.

[0083] The unit entering TH off (Thermal Off) state means the state of the compressor 2 stopping operation due to temperature or pressure protection mechanism.

[0084] The TH off state is the inherent safety mechanism of the air conditioning system. When the winding temperature of the compressor 2 or the system high pressure exceeds the safety threshold, the power supply of the compressor 2 will be automatically cut off to prevent equipment damage, and at this time the fan 3 still continues to run to assist heat dissipation.

[0085] When the unit enters the TH off state, the state must be maintained for at least 3 minutes before the compressor 2 is allowed to restart. Therefore, based on this unit characteristic, the second preset time period can be set to 3 minutes.

[0086] In some embodiments, the specific process of performing dirty block detection with the 3-minute compressor 2 downtime gap inherent in the TH off protection mechanism is as follows:

[0087] When the control unit 6 detects that the conditions for performing dirty block detection are met, the air conditioner operating mode is switched to a preset blowing mode.

[0088] After running for 3 minutes, which corresponds to the fan entering a stable blowing state, the fan motor power, indoor temperature, and indoor humidity data at this time are recorded as actual operating data.

[0089] After the collection is complete, the control unit 6 controls the unit to resume the operating mode before the blowing mode was switched.

[0090] Using the TH off protection mechanism for dirty block detection can minimize the impact on user experience:

[0091] Under the TH off state, the compressor 2 has stopped working while the fan 3 continues to run, and mode switching only adjusts the blowing parameters without interrupting the basic ventilation function.

[0092] The detection window is strictly limited to the 3-minute system pressure balance period required for safe restart of the compressor 2, without additional extension of downtime.

[0093] The mode recovery is completely synchronized with the timing of the compressor 2 restart, and actual measurements show that the user can only perceive changes at the moment of switching the deflector angle, and there is no temperature and humidity fluctuation.

[0094] In some embodiments, the above-mentioned preset blowing mode includes a specified air baffle and a specified deflector angle, and ensures that the selection of the specified air baffle and the specified deflector angle has little effect on user experience, i.e., little effect on environmental temperature and humidity fluctuations, without interfering with the collection of actual operating data.

[0095] In some embodiments, Figure 3 is a standard operating data set construction flowchart according to the embodiments of the present application, as Figure 3 shown, the standard operating data set construction method is as follows.

[0096] Obtain the operating history data of the air conditioner in the non-dirty block state, which includes the standard indoor temperature ti, the standard indoor humidity hi, and the standard fan power P0.

[0097] Generate n groups of standard working conditions based on clustering of the operating history data.

[0098] In some embodiments, K-means clustering analysis can be performed on the user historical operation data to extract typical working condition center points covering the actual use scenarios, forming n groups of standard working conditions.

[0099] For each group of standard working conditions, under the preset air supply mode, 1-m level dirty and blocked states are simulated respectively, and dirty and blocked fan power Pm(m=1, 2, …, m) is obtained. Based on the standard fan power P0 and the dirty and blocked fan power Pi, the multi-level dirty and blocked power difference {△P1, △P2, …, △Pm} is calculated and obtained.

[0100] Based on the operation history data and the multi-level dirty and blocked power difference, a standard operation data set is constructed, and the standard operation data set is denoted as {ti, hi, P0, △P1, △P2, …, △Pm}.

[0101] In some embodiments, the actual operation data includes a first indoor temperature, a first indoor humidity, and a first fan power.

[0102] Based on the weighted distance algorithm, the actual operation data is matched with the standard operation data set according to the first indoor temperature and the first indoor humidity, and the standard operation data corresponding to the closest standard working condition is obtained as the reference operation data. The reference operation data includes a second indoor temperature, a second indoor humidity, a second fan power, and a multi-level dirty and blocked power difference.

[0103] By dynamically matching the real-time temperature and humidity with the standard working condition library through the weighted distance algorithm, the optimal reference operation data is locked, the dirty and blocked misjudgment problem caused by environmental fluctuations is solved, and by using temperature and humidity as the judgment reference, the model generalization ability and the dirty and blocked recognition accuracy in complex use scenarios are improved.

[0104] In some embodiments, Figure 4 The standard fan power correction flowchart according to the embodiments of the present application is shown in FIG. 1. Figure 4 As shown in FIG. 1, when initially installed, the fan is controlled to operate in a preset air supply mode to stabilize under a standard working condition, and the real-time power value is recorded. According to the difference between the standard fan power corresponding to the standard working condition and the real-time power value, a compensation value is generated to correct all standard fan powers in the standard operation data set.

[0105] Specifically, since there are individual differences between different air conditioner motors, which can cause differences in operating power under the same working condition, power correction is needed when the air conditioner is initially installed.

[0106] Specifically, the correction method at initial installation is that the fan 3 is controlled to operate in the air supply mode (specified damper, specified angle of the air deflector) according to a standard working condition set in a stable manner, the real-time motor power value P' at this time is recorded, the difference between the standard fan power P0' corresponding to the standard working condition set without blockage and the real-time power value P' is calculated, and the difference is taken as a compensation value to correct the standard fan power values of all working conditions in the standard working condition set.

[0107] By calibrating the motor power reference value in real time at the initial installation stage, the error caused by individual differences of the motor is eliminated; the standard data set is corrected based on the compensation value, the dirty blockage judgment reference conforming to the actual air conditioner is constructed, the recognition results of different equipment under the same working condition remain high consistency, and the dirty blockage recognition accuracy is improved.

[0108] In some embodiments, a dual-path discriminant network model for performing dirty blockage recognition is constructed, as shown in Figure 5 The dual-path discriminant network model includes an actual running data feature extraction branch.

[0109] The actual running data feature extraction branch includes an input layer. After the actual running data is normalized, the actual running data feature extraction branch is input into the input layer.

[0110] Specifically, the actual running data includes the first indoor temperature, the first indoor humidity and the first fan power, and there is a large difference in the order of magnitude among the data, which is easy to cause gradient imbalance. Therefore, it is necessary to compress the actual running data to a unified interval to obtain the first indoor temperature normalized value, the first indoor humidity normalized value and the first fan power normalized value as the inputs of the actual running data feature extraction branch, so as to unify the gradient amplitude of each feature, improve the learning rate of the model and reduce the iteration number of the model.

[0111] The actual running data feature extraction branch includes a fully connected layer, which is used to reduce the calculation amount of the model to realize the lightweight of the model to adapt to the embedded demand.

[0112] Specifically, the fully connected layer is connected to the input layer, and a fully connected layer with 4 nodes is adopted. The fully connected layer maps the normalized input to a 4-dimensional feature space through a weight matrix, learns the nonlinear coupling relationship among the temperature, humidity and power under the premise of meeting the embedded resource constraint, realizes feature dimension reduction and key dirty blockage sensitive factor extraction, and maintains high-precision recognition ability of the model under low computing power condition.

[0113] The actual running data feature extraction branch includes a RELU activation function layer, which is used to apply nonlinear transformation to the features output by the fully connected layer, suppress negative value noise propagation, strengthen positively correlated dirty blockage sensitive signals and improve the proportion of effective feature weights.

[0114] Specifically, the RELU activation function layer is connected after the full connection layer, and the L2 regularization coefficient is set to 0.01 to prevent overfitting.

[0115] The actual operation data feature extraction branch includes a Dropout layer to enhance the environmental generalization ability of the model.

[0116] Specifically, the Dropout layer is connected after the RELU activation function layer, and a 4-dimensional actual feature vector is obtained through the Dropout layer. The Dropout layer actively introduces structural noise by randomly discarding 10% of the neurons in the training stage, forces the network to learn redundant feature expression, avoids over-reliance on a single neuron, enhances the environmental generalization ability of the model, and maintains stable performance of the model throughout the life cycle of the device.

[0117] In some embodiments, as shown in Figure 6 The two-way discriminant network model includes a benchmark operation data feature extraction branch.

[0118] The benchmark operation data feature extraction branch includes an input layer. After the benchmark operation data is normalized, the benchmark operation data is input into the input layer of the benchmark operation data feature extraction branch.

[0119] Specifically, the benchmark operation data includes the second indoor temperature, the second indoor humidity, the second fan power, and the multi-level dirty block power difference. There is a large difference in the order of magnitude between the data, which can easily cause gradient imbalance. Therefore, it is necessary to compress the actual operation data to a unified interval to obtain the second indoor temperature normalized value, the second indoor humidity normalized value, the second fan power normalized value, and the multi-level dirty block power difference normalized value as the input of the benchmark operation data feature extraction branch, so that the gradient amplitude of each feature is unified, the learning rate of the model is improved, and the number of model iteration rounds is reduced.

[0120] In some embodiments, taking the multi-level dirty block power difference set to 7 levels as an example, that is, the multi-level dirty block power difference is recorded as {△P1, △P2, …, △P7}, and the input layer of the benchmark operation data feature extraction branch has 10 nodes.

[0121] The benchmark operation data feature extraction branch includes a full connection layer to reduce the calculation amount of the model and realize the lightweight of the model to adapt to the embedded requirements.

[0122] Specifically, the full connection layer is connected after the input layer, and a full connection layer with 6 nodes is adopted. Since the benchmark operation data feature extraction branch needs to process higher-dimensional features, that is, 10-dimensional features, the 6-node full connection layer can provide sufficient feature abstraction capability.

[0123] The full connection layer maps the normalized input to a 6-dimensional feature space through a weight matrix, learns the nonlinear coupling relationship between temperature, humidity, power, and multi-level dirty and blocked power difference under the premise of meeting the embedded resource constraints, realizes feature dimension reduction and key dirty and blocked sensitive factor extraction, and enables the model to maintain high-precision recognition ability under low computing power conditions.

[0124] The reference operation data feature extraction branch includes a RELU activation function layer, which is used to apply a nonlinear transformation to the features output by the full connection layer, suppress negative noise propagation, strengthen positively correlated dirty and blocked sensitive signals, and improve the proportion of effective feature weights.

[0125] Specifically, the RELU activation function layer is connected to the full connection layer, and the L2 regularization coefficient is set to 0.01 to prevent overfitting.

[0126] The reference operation data feature extraction branch includes a Dropout layer to enhance the environmental generalization ability of the model.

[0127] Specifically, the Dropout layer is connected to the RELU activation function layer, and a 6-dimensional reference feature vector is obtained through the Dropout layer. The Dropout layer randomly discards 10% of the neurons during the training phase to enhance the environmental generalization ability of the model, so that the model maintains stable performance throughout the device life cycle.

[0128] In some embodiments, as shown in FIG. 1, Figure 7 The dual-path discriminative network model includes a dual-path feature fusion branch.

[0129] The dual-path feature fusion branch includes a Concat connection layer to concatenate the actual feature vector and the reference feature vector to obtain a feature concatenation vector. The Concat connection layer can output lossless fused dual-path features to avoid information loss caused by dimension reduction.

[0130] Specifically, the 4-dimensional actual feature vector and the 6-dimensional reference feature vector are concatenated to obtain a 10-dimensional feature concatenation vector.

[0131] The dual-path feature fusion branch includes a first full connection layer that performs cross-modal feature dimension reduction on the feature concatenation vector based on a weight matrix.

[0132] Specifically, the first full connection layer is connected to the Concat connection layer and uses an 8-node full connection layer to reduce the 10-dimensional feature concatenation vector to an 8-dimensional feature concatenation vector.

[0133] The dual-path feature fusion branch includes a RELU activation function layer connected to the first full connection layer to apply a nonlinear transformation to the features output by the first full connection layer, suppress negative noise propagation, strengthen positively correlated dirty and blocked sensitive signals, and improve the proportion of effective feature weights.

[0134] The double-path feature fusion branch includes a second fully connected layer, which maps the abstract features to a dimensional space consistent with the dirty block level nodes and learns the discriminant weight of each level.

[0135] Specifically, the second fully connected layer is connected to the RELU activation function layer, and taking a 7-level dirty block power difference setting as an example, the second connected layer correspondingly adopts 7 nodes to map the abstract features output by the RELU activation function layer to a 7-dimensional space consistent with the dirty block level nodes, thereby providing a dimensionally matched input for Softmax.

[0136] The double-path feature fusion branch includes a Softmax activation function layer, which quantifies the severity of the dirty block in the form of a probability value, and outputs the probability distribution of the dirty block level of the air conditioner through the Softmax activation function layer to obtain the dirty block level.

[0137] Specifically, taking a 7-level dirty block power difference setting as an example, the expression of the Softmax activation layer function is:

[0138]

[0139] wherein, is the predicted probability of the mthlevel dirty block; S m is the original classification score corresponding to the mthlevel dirty block; is an exponential operation on S m ; is the sum of the exponential scores of all dirty block levels.

[0140] The probability representation of the dirty block level of the air conditioner is: wherein the dirty block level with the highest probability is the dirty block level of the unit.

[0141] It should be noted that the levels of the multi-level dirty block power difference are not limited to 7 levels. The fan power base of different capacity models is different, and the power fluctuation amplitude caused by the dirty block is different. In addition, the air duct resistance characteristics of different model structures are different, and the influence of the dirty block on the power is different. The number of levels of the multi-level dirty block power difference can be adjusted according to the above differences.

[0142] It should be noted that the above double-path discriminant network model is only one embodiment, and the node settings, activation function types, regularization coefficients, Dropout rates and other hyperparameters of each layer can be dynamically adjusted according to the hardware performance of the model, the calculation resource limitation and the dirty block recognition accuracy requirement.

[0143] In some embodiments, the double-path discriminant network model as shown in Figure 8 can be trained using a cross-entropy loss function combined with an adam optimizer.

[0144] Specifically, the learning rate is set to 1e-4, the batch size is set to 256, and the training round is set to 50. The iterative optimization of the model parameters is realized by the above parameter configuration to improve the dirty block level recognition ability.

[0145] After the model training is completed, verification is needed. The specific process is to input the pre-reserved verification set data into the trained model, predict the dirty block level of the verification set samples through the model, and then calculate the proportion of the number of samples with prediction accuracy to the total number of samples in the verification set as the standard for evaluating the performance of the model.

[0146] In some embodiments, taking a four-direction machine as an example, parameter determination can be performed under the set test conditions according to Table 1 to obtain a double-path discrimination network model training verification set.

[0147]

[0148]

[0149] Table 1 Test conditions for motor power determination of four-direction machines under different working conditions

[0150] It should be noted that the test conditions are not limited to the working conditions listed in the above table. The working conditions can be expanded from multiple dimensions such as operating mode, environmental parameters, and machine type differences to cover more comprehensive data.

[0151] In some embodiments, if the dirty block level reaches a preset level, the control unit 6 controls the air conditioner to issue an alarm display.

[0152] By triggering the intelligent alarm mechanism through the preset dirty block level threshold, real-time alarm of moderate to severe dirty block that needs to be intervened is realized, so that the user can respond to the dirty block situation in a timely manner, effectively avoiding the energy consumption surge and equipment damage caused by the dirty block problem.

[0153] In some embodiments, Figure 9 is a module linkage flowchart according to the embodiments of the present application, as Figure 9 shown, based on whether the dirty block level exceeds the preset level, it is determined whether dirty block site judgment is needed. If the preset level is exceeded, further dirty block site judgment is performed. If the dirty block site is the filter screen, the cloud is used to remind the user through the APP end, and the user is advised to clean the air conditioner filter screen in a timely manner. If the dirty block site is the heat exchanger, the self-cleaning function is enabled to realize the logic, and the current environment is identified to find the appropriate time to run the self-cleaning function.

[0154] The embodiments of the present application also provide an air conditioner control method.

[0155] In response to the first preset time period, it is determined whether the air conditioner is in the state of starting and the compressor 2 stopping;

[0156] If the start-up and compressor 2 stop state is met, the air conditioner is controlled to run in a preset air supply mode for a second preset time length, and actual running data is obtained;

[0157] Based on the actual running data and a pre-determined standard running data set, the closest standard running data corresponding to a standard working condition is matched and obtained as reference running data, wherein the reference running data includes a multi-grade dirty and blocked power difference;

[0158] The actual running data and the reference running data are input into a double-path discriminant network model, feature extraction is performed respectively, feature fusion is performed, feature recognition is performed through the double-path discriminant network model, a probability distribution of each dirty and blocked grade is output according to the multi-grade dirty and blocked power difference, and a dirty and blocked grade of the air conditioner is obtained.

[0159] In addition, in combination with Figure 1 The air conditioner described in the embodiments of the present application. Figure 10 A hardware configuration diagram of the control unit 6 according to the embodiments of the present application.

[0160] The control unit 6 of the air conditioner can include a processor 81 and a memory 82 storing computer program instructions.

[0161] Specifically, the processor 81 described above can include a central processing unit 81 (CPU), or an application specific integrated circuit (ASIC), or can be configured as one or more integrated circuits that implement the embodiments of the present application.

[0162] The memory 82 can include a mass storage device 82 for data or instructions.

[0163] In some embodiments, the control unit 6 of the air conditioner can further include a communication interface 83 and a bus 80. As shown in Figure 10 The processor 81, the memory 82, and the communication interface 83 are connected through the bus 80 and complete communication with each other.

[0164] The communication interface 83 is used to realize communication between modules, devices, units and / or equipment in the embodiments of the present application.

[0165] The bus 80 includes hardware, software or both, which couples components of the air conditioner to each other.

[0166] The technical features of the above-described embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described, however, as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present disclosure.

[0167] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and in a specific and detailed manner, but should not be construed as limiting the scope of the patent. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. An air conditioner, characterized in that: It includes: The casing has an air outlet and a return air inlet provided on its periphery; A compressor and a fan are arranged in the casing; a control unit electrically connected to the compressor and the fan; The control unit is configured to: In response to a first preset time, determining whether the air conditioner is in an on-state and the compressor is in a stopped state; If the power-on state is met and the compressor is stopped, controlling the air conditioner to operate in a preset air supply mode for a second preset time period, and obtaining actual operating data; Based on the actual operating data and a predetermined standard operating data set, matching and obtaining the standard operating data corresponding to the closest standard operating condition as the benchmark operating data, wherein the benchmark operating data includes multi-level dirty and blocked power differences; The actual operation data and the benchmark operation data are input into a two-way discriminant network model, and feature extraction and feature fusion are performed on each of them. Feature recognition is performed by the two-way discriminant network model, and the probability distribution of each dirty clogging level is output based on the multi-level dirty clogging power difference to obtain the dirty clogging level of the air conditioner.

2. The air conditioner according to claim 1, characterized in that The control unit is further configured to: Acquire operation history data of the air conditioner in a non-clogging state, wherein the operation history data includes a standard indoor temperature ti, a standard indoor humidity hi, and a standard fan power P0; Based on the clustering of the operation history data, n groups of standard operating conditions are generated. For each group of standard operating conditions, under a preset air supply mode, dirty blockage states of levels 1 to m are simulated and dirty blockage fan powers Pm (m=1, 2, ..., m) are obtained. Based on the standard fan power P0 and the dirty blockage fan power Pi, multi-level dirty blockage power differences {△P1, △P2, ..., △Pm} are calculated; A standard operation data set is constructed based on the operation history data and the multi-level dirty and congestion power differences, and the standard operation data set is recorded as {ti, hi, P0, ΔP1, ΔP2, ..., ΔPm}.

3. The air conditioner according to claim 2, characterized in that The control unit is further configured to: The actual operation data includes a first indoor temperature, a first indoor humidity and a first fan power; Based on a weighted distance algorithm, the actual operating data is matched with the standard operating data set according to the first indoor temperature and the first indoor humidity, and the standard operating data corresponding to the closest standard operating condition is obtained as the benchmark operating data. The benchmark operating data includes the second indoor temperature, the second indoor humidity, the second fan power, and a multi-level dirty and clogging power difference.

4. The air conditioner according to claim 3, characterized in that The control unit is further configured to: During initial installation, under a standard operating condition, the fan is controlled to operate in a preset air supply mode until it is stable, and the real-time power value is recorded; A compensation value is generated according to the difference between the standard fan power corresponding to the standard operating condition and the real-time power value, and all standard fan powers in the standard operation data set are corrected.

5. The air conditioner according to claim 1, characterized in that The control unit is further configured to: The dual-path discriminant network model includes an actual operation data feature extraction branch and a reference operation data feature extraction branch; Input the actual operation data into the input layer of the actual operation data feature extraction branch, and process it through the fully connected layer, the ReLU activation function layer and the Dropout layer in sequence to obtain the actual feature vector; The benchmark operation data is input into the input layer of the benchmark operation data feature extraction branch, and is processed in sequence by a fully connected layer, a ReLU activation function layer, and a Dropout layer to obtain a benchmark feature vector.

6. The air conditioner according to claim 5, characterized in that The control unit is further configured to: After normalization processing is performed on the actual operation data and the reference operation data, the data are input into the actual operation data feature extraction branch and the reference operation data feature extraction branch respectively.

7. The air conditioner according to claim 5, characterized in that: The dual-path discriminant network model also includes a dual-path feature fusion branch; The actual feature vector and the reference feature vector are input into the dual-path feature fusion branch for feature splicing, and then sequentially passed through the first fully connected layer, the RELU activation function layer, the second fully connected layer and the Softmax activation function to output the probability distribution of the dirtiness and blockage level of the air conditioner to obtain the dirtiness and blockage level.

8. The air conditioner according to claim 1, wherein: The control unit is further configured to control the air conditioner to issue an alarm display if the dirt and blockage level reaches a preset level.

9. The air conditioner according to claim 1, wherein: The control unit is further configured to: The control unit controls the air conditioner to restore a preset operation mode after acquiring the actual operation data.

10. An air conditioner control method, characterized in that: include: In response to a first preset time, determining whether the air conditioner is in an on-state and the compressor is in a stopped state; If the power-on state is met and the compressor is stopped, controlling the air conditioner to operate in a preset air supply mode for a second preset time period, and obtaining actual operating data; Based on the actual operating data and a predetermined standard operating data set, matching and obtaining the standard operating data corresponding to the closest standard operating condition as the benchmark operating data, wherein the benchmark operating data includes multi-level dirty and blocked power differences; The actual operation data and the benchmark operation data are input into a two-way discriminant network model, and feature extraction and feature fusion are performed on each of them. Feature recognition is performed by the two-way discriminant network model, and the probability distribution of each dirty clogging level is output based on the multi-level dirty clogging power difference to obtain the dirty clogging level of the air conditioner.

Images