Air conditioner

Abstract

The application discloses an air conditioner, which comprises a motor, a driving module, a voltage detection module, a current detection module, a temperature detection module and a controller. The driving module is used for driving the motor to operate, and the driving module comprises at least one intelligent power module. A junction temperature prediction model is configured in the controller, and the controller is configured to obtain a junction temperature prediction value of the intelligent power module through the junction temperature prediction model, generate a junction temperature dynamic threshold value of the intelligent power module according to a measured ambient temperature and a motor current, and execute a target protection strategy for the intelligent power module according to a temperature difference condition of the junction temperature prediction value and the junction temperature dynamic threshold value. The air conditioner can predict the junction temperature of the intelligent power module in advance through the junction temperature prediction model, thereby quickly and accurately obtaining the junction temperature prediction value, and the intelligent power module can be protected according to the temperature difference condition, so that the protection effect of the intelligent power module is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of air conditioners, in particular to an air conditioner. BACKGROUND

[0002] In the prior art, when measuring the junction temperature of the intelligent power module in the actual use of the air conditioner, the traditional temperature sensor will produce a certain delay due to the long heat conduction path, and cannot timely capture the mutation of the junction temperature of the intelligent power module chip, resulting in lag of protection action; when the existing junction temperature prediction model predicts the junction temperature, the input parameters are only the parameters of the intelligent power module itself, resulting in large error of the predicted value of the junction temperature of the intelligent power module, and the protection mode of the intelligent power module is single, which is easy to cause damage to the intelligent power module. SUMMARY

[0003] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, one object of the present application is to provide an air conditioner which can predict the junction temperature of the intelligent power module in advance through a junction temperature prediction model, so as to quickly and accurately obtain the predicted value of the junction temperature, and protect the intelligent power module according to the temperature difference condition, thereby improving the protection effect of the intelligent power module.

[0004] In order to solve the above problems, the first aspect of the present application provides an air conditioner, comprising: a motor and a drive module, the drive module being used to drive the motor to operate, the drive module comprising at least one intelligent power module; a voltage detection module, used to detect the DC bus voltage of a DC bus connected with the drive module; a current detection module, used to detect the motor phase current or the DC bus current; a temperature detection module, used to detect the ambient temperature; a controller, connected with the drive module, the voltage detection module, the current detection module and the temperature detection module, the controller being configured with a junction temperature prediction model, the controller being configured to: obtain the predicted value of the junction temperature of the intelligent power module through the junction temperature prediction model, wherein the junction temperature prediction model takes at least the DC bus voltage, the motor phase current or the DC bus current and the ambient temperature as input, and the junction temperature prediction model takes the predicted value of the junction temperature of the intelligent power module as output; generate the dynamic threshold value of the junction temperature of the intelligent power module according to the measured ambient temperature and the motor current; and execute the target protection strategy for the intelligent power module according to the temperature difference condition of the predicted value of the junction temperature and the dynamic threshold value of the junction temperature.

[0005] According to the air conditioner provided in the embodiment of the present application, the junction temperature prediction model takes at least the DC bus voltage, the motor phase current or the DC bus current and the measured ambient temperature as input parameters, that is, the influence of more parameters on the junction temperature of the intelligent power module is considered, the junction temperature value of the intelligent power module is predicted according to a larger amount of input parameters, the error of the junction temperature prediction value of the intelligent power module can be effectively reduced, and the accuracy of the junction temperature prediction value is improved. Moreover, the junction temperature dynamic threshold of the intelligent power module is obtained according to the measured ambient temperature and the motor current, the junction temperature prediction value of the intelligent power module output by the junction temperature prediction model is compared with the junction temperature dynamic threshold to obtain a temperature difference condition, and the intelligent power module is protected according to the temperature difference condition, wherein the measured ambient temperature and the motor circuit both change, so that the obtained junction temperature dynamic threshold also dynamically changes, compared with the fixed threshold, the temperature difference can be dynamically judged based on the actual situation, the accuracy of the temperature difference condition judgment is improved, so that a more effective protection strategy can be executed, and the protection effect on the intelligent power module is improved.

[0006] In some embodiments, the junction temperature prediction model takes the DC bus voltage, the motor phase current or the DC bus current, the ambient temperature, the switching frequency of the intelligent power module and the motor rotating speed as inputs.

[0007] The above technical solution has the following advantages or beneficial effects: the input parameters of the existing junction temperature prediction model are only the parameters of the power module itself, the DC bus voltage, the motor phase current or the DC bus current, the ambient temperature, the switching frequency of the intelligent power module and the motor rotating speed are taken as inputs in the junction temperature prediction model, the junction temperature value of the intelligent power module is predicted through a large amount of input parameters, and the error of the junction temperature prediction value of the intelligent power module is effectively reduced.

[0008] In some embodiments, the DC bus voltage is a voltage signal subjected to preset voltage amplitude fluctuation processing on an original DC bus voltage.

[0009] The above technical solution has the following advantages or beneficial effects: for the use environment of the air conditioner, when the input power grid has abnormal fluctuation, the phenomenon of abnormal rise of the DC bus voltage is prone to occur, the original DC bus voltage is subjected to preset voltage amplitude fluctuation processing to simulate abnormal fluctuation of the power grid, and the junction temperature prediction value of the intelligent power module output by the junction temperature prediction model is more accurate.

[0010] In some embodiments, the controller is further configured to obtain the junction temperature dynamic threshold through a junction temperature dynamic threshold model, wherein the junction temperature dynamic threshold model takes the measured ambient temperature and the motor current as inputs.

[0011] The technical scheme has the following advantages or beneficial effects: the junction temperature dynamic threshold model is used to obtain the junction temperature dynamic threshold of the intelligent power module under the current condition, the junction temperature dynamic threshold is different under different input parameters, and the intelligent power module is protected according to the different junction temperature dynamic thresholds, thereby reducing the failure rate.

[0012] In some embodiments, the junction temperature dynamic threshold model is expressed as follows:

[0013] Tj_threshold = Tj_base + K1×(Tamb-25) + K2×(Iavg / Irated);

[0014] wherein Tj_threshold is the junction temperature dynamic threshold, Tj_base is the temperature safety threshold of the intelligent power module, Tj_base = 125℃, K1 is an environmental compensation coefficient, Tamb is the measured ambient temperature, K2 is a motor load rate compensation coefficient, Iavg is the motor average current, and Irated is the motor rated current.

[0015] The technical scheme has the following advantages or beneficial effects: the environmental compensation coefficient, the measured ambient temperature, the motor load rate compensation coefficient, the motor average current and the motor rated current are input into the junction temperature dynamic threshold model to obtain the junction temperature dynamic threshold of the intelligent power module under the current condition, and the intelligent power module is protected according to the junction temperature dynamic threshold, thereby reducing the failure rate.

[0016] In some embodiments, the controller implements a hardware-software cooperative protection strategy and is configured to perform at least one of the following when executing a target protection strategy for the intelligent power module according to the temperature difference condition: when the temperature difference condition is Tj_pre ≥ Tj_threshold + 5℃, the target protection strategy is to perform fault alarm and immediately shut off the control signal of the intelligent power module, wherein Tj_pre is a junction temperature prediction value and Tj_threshold is the junction temperature dynamic threshold; when the temperature difference condition is Tj_threshold ≤ Tj_pre < Tj_threshold + 5℃, the target protection strategy is to perform fault alarm and the intelligent power module is reduced to 50% and starts forced heat dissipation; when the temperature difference condition is Tj_threshold - 10℃ ≤ Tj_pre < Tj_threshold, the target protection strategy is to trigger the PWM (Pulse Width Modulation) duty cycle linear reduction of the intelligent power module control signal; and when the temperature difference condition is Tj_pre < Tj_threshold - 10℃, the target protection strategy is to allow the intelligent power module to run at an over frequency (maximum + 10% switching frequency).

[0017] The technical scheme has the advantages or beneficial effects that the junction temperature prediction value is compared with the junction temperature dynamic threshold to obtain a temperature difference condition, the intelligent power module is adjusted according to the temperature difference condition, different processing measures are taken according to the degree of abnormal temperature rise of the intelligent power module, efficient and rapid high-temperature early warning and protection are realized, and abnormal damage of the machine is avoided.

[0018] In some embodiments, the junction temperature prediction model is a recurrent neural network model with a forgetting gate mechanism.

[0019] The technical scheme has the advantages or beneficial effects that the forgetting gate mechanism enables the recurrent neural network model to dynamically adjust the retention and forgetting of information in the cell state, thereby effectively capturing long-term dependencies in long sequences. When processing long sequence data, the long recurrent neural network model can selectively forget unimportant historical information through the forgetting gate while retaining information useful for the current task.

[0020] In some embodiments, the recurrent neural network model with a forgetting gate mechanism includes one of a long short-term memory network model and a network model derived based on the long short-term memory network model.

[0021] The technical scheme has the advantages or beneficial effects that the recurrent neural network model with a forgetting gate mechanism includes one of a long short-term memory network model and a network model derived based on the long short-term memory network model, and a large amount of data in and out can be learned through the above model, thereby accurately and quickly outputting the junction temperature prediction value of the intelligent power module.

[0022] In some embodiments, the long short-term memory network model includes an input layer, a hidden layer, and an output layer, wherein the input layer is configured as an input parameter of a preset time step, the hidden layer includes a double-layer long short-term memory network, a first layer long short-term memory network is used to extract time sequence features, and a second layer long short-term memory network is used to capture long-period cumulative effects, a preset Dropout value is added between the first layer long short-term memory network and the second layer long short-term memory network, and the output layer is used to output the junction temperature prediction value of the intelligent power module; a loss function of the long short-term memory network model adopts Huber loss, and an optimizer of the long short-term memory network model adopts an Adam algorithm.

[0023] The above technical scheme has the following advantages or beneficial effects: the long short-term memory network is a powerful recurrent neural network model, by introducing a gating mechanism, long-term dependencies in long sequences can be effectively captured and remembered, by stacking multiple hidden layers to build a deeper network structure, the complexity and expression ability of the model are improved, so that the junction temperature prediction value of the intelligent power module is accurately output, and by adopting the Huber loss and the Adam algorithm, the error can be reduced and the efficiency can be improved.

[0024] In some embodiments, the junction temperature prediction value of the intelligent power module is a junction temperature peak value of the intelligent power module within a preset time period in the future.

[0025] The above technical scheme has the following advantages or beneficial effects: the junction temperature peak value of the intelligent power module within a preset time period in the future is predicted by the junction temperature prediction model, so that the intelligent power module is protected in advance according to the junction temperature peak value, and the failure rate is reduced.

[0026] Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following and / or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0027] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0028] Figure 1 is a perspective view of an air conditioner according to an embodiment of the present application;

[0029] Figure 2 is a structural block diagram of an air conditioner according to an embodiment of the present application;

[0030] Figure 3 is a structural block diagram of a drive module according to an embodiment of the present application;

[0031] Figure 4 is a structural block diagram of an air conditioner according to an embodiment of the present application;

[0032] Figure 5 is a control flowchart of a controller according to an embodiment of the present application;

[0033] Figure 6 is a schematic diagram of junction temperature prediction according to an embodiment of the present application;

[0034] Figure 7 is a schematic diagram of a long short-term memory network model structure according to an embodiment of the present application;

[0035] Figure 8 is a schematic diagram of a long short-term memory network unit according to an embodiment of the present application.

[0036] Reference signs:

[0037] Air conditioner 100

[0038] Outdoor unit 1; connecting pipe 2; indoor unit 3; motor 110; drive module 120; voltage detection module 130; current detection module 140; temperature detection module 150; controller 160; intelligent power module 121. DETAILED DESCRIPTION

[0039] The embodiments of the present application are described in detail below, and the embodiments described with reference to the accompanying drawings are exemplary, and the embodiments of the present application are described in detail below.

[0040] The air conditioner realizes the refrigeration and heating cycle of the air conditioner through the refrigerant circulation system.

[0041] The refrigerant circulation system includes a compressor, which is usually located in the air conditioner outdoor unit. The compressor is a device that converts low-temperature and low-pressure gaseous refrigerant into high-temperature and high-pressure refrigerant through work. It is the "heart" of the air conditioning system, mainly responsible for compressing and transporting refrigerant to achieve the function of refrigeration or heating. In the refrigeration cycle, the compressor extracts the refrigerant from the low-pressure area, sends it to the high-pressure area after compression, and then cools and condenses it. After the pressure is reduced by the throttling device, it enters the evaporator to evaporate and absorb heat, thereby adjusting the temperature, humidity, and other parameters in the room or vehicle cabin.

[0042] The refrigerant circulation system also includes a throttling device, which can be a capillary tube or an expansion valve. Taking the expansion valve as an example, the expansion valve changes the high-temperature and high-pressure liquid refrigerant into low-temperature and low-pressure wet steam through throttling to achieve the refrigeration effect. The expansion valve is mainly installed between the liquid storage tank and the evaporator, and the valve flow is controlled by the heat change at the end of the evaporator, thereby preventing the occurrence of insufficient evaporator area utilization and knocking cylinder phenomenon.

[0043] The refrigerant circulation system also includes a switching device, which can be a four-way valve. The four-way valve has four channels or connections and is a key component in the refrigeration and air conditioning system for changing the flow direction of the refrigerant. The four-way valve can realize the switching between the refrigeration and heating conditions of the air conditioner by changing the flow direction of the refrigerant. The four-way valve is mainly composed of an electromagnetic directional valve and a four-way reversing valve. The electromagnetic directional valve is composed of a valve bowl, a spring, an iron core, and an electromagnetic coil, while the four-way reversing valve is controlled by the electromagnetic directional valve, and the two are connected by a directional capillary tube.

[0044] The refrigerant circulation system also includes an evaporator, which is specifically designed for the liquid refrigerant to boil and evaporate. The evaporation of the liquid refrigerant absorbs heat from the room, lowering the room temperature, which is the ultimate manifestation of its cooling capacity. During the process of absorbing heat and evaporating to cool the air, the evaporator causes water vapor in the air to condense and separate on the surface of the coils, reducing air humidity and thus having a dehumidifying effect.

[0045] The refrigerant circulation system also includes a condenser passage. As one of the core components of the refrigeration system, the condenser passage efficiently completes the heat exchange task, ensuring the system maintains stable performance during continuous operation. Through its piping design, the condenser passage allows high-temperature, high-pressure refrigerant vapor to enter and exchange heat with the external environment (such as air or water). In the passage, the refrigerant releases a large amount of heat, its temperature gradually decreases, and it eventually changes from a gaseous state to a liquid state. This process achieves effective heat transfer and dissipation, and is a crucial element for the continuous operation of the refrigeration system.

[0046] The outdoor unit of an air conditioner refers to the part of the refrigeration cycle that includes the compressor and the outdoor heat exchanger. The indoor unit of an air conditioner includes the indoor heat exchanger, and the expansion valve can be provided in either the indoor or outdoor unit.

[0047] The indoor and outdoor heat exchangers function as either condensers or evaporators. When the indoor heat exchanger is used as a condenser, the air conditioner functions as a heater in heating mode; when the indoor heat exchanger is used as an evaporator, the air conditioner functions as a cooler in cooling mode.

[0048] like Figure 1 As shown, the air conditioner 100 includes an indoor unit 3 and an outdoor unit 1. The indoor unit 3 is usually installed indoors, and the outdoor unit 1 is usually installed outdoors for heat exchange in the indoor environment. The outdoor unit 1 and the indoor unit 3 are connected by a connecting pipe 2 to perform cooling and heating.

[0049] In existing technologies, air conditioner outdoor unit drive modules incorporate intelligent power modules. When measuring the junction temperature of these modules under actual air conditioner usage, traditional temperature sensors experience a delay due to long heat conduction paths, failing to promptly capture sudden changes in junction temperature and resulting in delayed protection actions. Furthermore, in the air conditioner's operating environment, abnormal fluctuations in the input power grid can easily lead to abnormal increases in DC bus voltage, causing the intelligent power module to overheat, potentially damaging the module or causing abnormal shutdown. Linear models based on temperature rise formulas cannot account for nonlinear factors such as power grid harmonics and alternating ambient temperature and humidity, resulting in significant errors in junction temperature estimation. Existing junction temperature prediction models only input the intelligent power module's own parameters, leading to large errors in predicted junction temperature values. Moreover, they lack tiered protection for the intelligent power module, resulting in a single protection method that easily damages the module or causes abnormal air conditioner shutdown, leading to high maintenance costs.

[0050] To solve the above problems, an air conditioner is provided in an embodiment of the first aspect of the present application. The air conditioner can predict the junction temperature of an intelligent power module in advance through a junction temperature prediction model, thereby quickly and accurately obtaining a junction temperature prediction value, and protecting the intelligent power module according to a temperature difference condition, thereby improving the protection effect on the intelligent power module.

[0051] As shown in Figure 2 , the air conditioner 100 comprises a motor 110, a drive module 120, a voltage detection module 130, a current detection module 140, a temperature detection module 150, and a controller 160.

[0052] The drive module 120 is configured to drive the motor 110 to operate. As shown in Figure 3 , the drive module 120 comprises at least one intelligent power module 121; the voltage detection module 130 is configured to detect the DC bus voltage of a DC bus connected to the drive module 120; the current detection module 140 is configured to detect the phase current of the motor 110 or the DC bus current; the temperature detection module 150 is configured to detect the ambient temperature; and the controller 160 is connected to the drive module 120, the voltage detection module 130, the current detection module 140, and the temperature detection module 150, and a junction temperature prediction model is configured in the controller 160.

[0053] Specifically, the air conditioner 100 comprises the motor 110 and the drive module 120. The motor 110 can be the motor of the outdoor fan of the air conditioner 100, or the motor of the outdoor compressor of the air conditioner 100. The drive module 120 is configured to drive the motor 110. As shown in Figure 2 , the drive module 120 can be arranged outside the motor 110, or as shown in Figure 4 , the drive module 120 can be integrated inside the motor 110.

[0054] The intelligent power module 121 combines power electronics and integrated circuit technology, integrating power switching devices (such as IGBTs and MOSFETs) with the drive module 120, making the drive module 120 more compact. Integrating power switching devices and drive circuits into a single module reduces size and development time. The intelligent power module 121 has built-in fault detection circuitry, such as overvoltage, overcurrent, and overheat protection, and can send detection signals to the control module 160. The intelligent power module 121 controls the IGBT's turn-on and turn-off through its internal drive circuitry. When the intelligent power module 121 receives a control signal, the drive circuit generates a corresponding gate voltage, causing the IGBT to turn on or off. Simultaneously, the internal protection circuitry of the intelligent power module 121 monitors the IGBT's operating status in real time. Once abnormal conditions such as overcurrent, overvoltage, or overheating are detected, the protection circuitry immediately activates, turning off the IGBT and outputting a fault signal.

[0055] Junction temperature refers to the operating temperature of the PN junction inside a semiconductor chip. It is a critical parameter for the smart power module 121, directly affecting its performance and lifespan. Excessive junction temperature can alter the properties of the semiconductor material, affecting its normal operation and potentially causing permanent damage. Therefore, the controller 160 incorporates a junction temperature prediction model to predict the junction temperature of the smart power module 121. When the junction temperature is too high, appropriate protection measures are implemented to prevent damage to the smart power module 121.

[0056] Based on the architecture of the air conditioner described above, refer to Figure 5 As shown, the controller of the air conditioner is configured to perform the following steps S1-S3.

[0057] Step S1: Obtain the predicted junction temperature of the smart power module through the junction temperature prediction model.

[0058] The junction temperature prediction model takes at least the DC bus voltage, motor phase current or DC bus current and ambient temperature as inputs and outputs the junction temperature prediction value of the intelligent power module.

[0059] Specifically, a junction temperature prediction model is configured in the controller 160, the junction temperature of the intelligent power module 121 is predicted by the junction temperature prediction model, the junction temperature prediction value of the intelligent power module 121 is obtained, when the junction temperature prediction value of the intelligent power module 121 is too high, corresponding protection measures are taken for the intelligent power module 121, and damage to the intelligent power module 121 is avoided. When predicting the junction temperature of the intelligent power module 121, the voltage detection module 130 sends the detected DC bus voltage of the DC bus connected to the drive module 120 to the controller 160, the current detection module 140 sends the detected motor phase current or DC bus current to the controller 160, and the temperature detection module 150 sends the detected environmental temperature to the controller 160; the junction temperature prediction model in the controller 160 takes at least the DC bus voltage, the motor phase current or the DC bus current and the environmental temperature as input, predicts the junction temperature of the intelligent power module 121, and outputs the junction temperature prediction value of the intelligent power module 121.

[0060] Step S2, generating a junction temperature dynamic threshold of the intelligent power module according to the measured environmental temperature and the motor current.

[0061] Specifically, the junction temperature dynamic threshold is generated according to the measured environmental temperature and the motor current, and since the environmental temperature and the motor current detected at different times are different, the junction temperature threshold is a dynamic threshold; the measured environmental temperature is the current environmental temperature detected by the temperature detection module, and the motor current reflects the actual working current condition of the equipment in a period of time. In the present application, the motor current can be the current of the fan motor of the outdoor unit of the air conditioner, or the current of the compressor motor of the outdoor unit of the air conditioner.

[0062] Step S3, executing a target protection strategy for the intelligent power module according to the temperature difference condition of the junction temperature prediction value and the junction temperature dynamic threshold.

[0063] Specifically, after the junction temperature prediction model outputs the junction temperature prediction value of the intelligent power module 121, the junction temperature prediction value is compared with the junction temperature dynamic threshold, so as to determine the temperature difference condition, which can be understood as a condition determined according to the junction temperature prediction value and the junction temperature dynamic threshold, and is used to judge the temperature difference between the junction temperature prediction value and the junction temperature dynamic threshold. According to the temperature difference condition of the junction temperature prediction value and the junction temperature dynamic threshold, corresponding protection measures are taken for the intelligent power module 121 to avoid damage to the intelligent power module 121; since different temperature difference conditions correspond to different protection strategies, after obtaining the temperature difference condition satisfied by the junction temperature prediction value and the junction temperature dynamic threshold, the target protection strategy is determined according to the temperature difference condition to protect the intelligent power module 121, and the target protection strategy can be understood as a strategy selected from multiple protection strategies and currently needed to protect the intelligent power module 121.

[0064] According to the air conditioner provided in the embodiment of the present application, the junction temperature prediction model takes at least the DC bus voltage, the motor phase current or the DC bus current and the measured ambient temperature as input parameters, that is, the influence of more parameters on the junction temperature of the intelligent power module is considered, the junction temperature value of the intelligent power module is predicted according to a larger amount of input parameters, the error of the junction temperature prediction value of the intelligent power module can be effectively reduced, and the accuracy of the junction temperature prediction value is improved. Moreover, the junction temperature dynamic threshold of the intelligent power module is obtained according to the measured ambient temperature and the motor current, the junction temperature difference condition is obtained by comparing the junction temperature prediction value of the intelligent power module output by the junction temperature prediction model with the junction temperature dynamic threshold, the intelligent power module is protected according to the junction temperature difference condition, wherein the measured ambient temperature and the motor circuit both change, so that the obtained junction temperature dynamic threshold also dynamically changes, compared with the fixed threshold, the junction temperature difference can be dynamically judged based on the actual situation, the accuracy of the junction temperature difference condition judgment is improved, so that a more effective protection strategy can be executed, and the protection effect of the intelligent power module is improved.

[0065] In some embodiments, the junction temperature prediction model takes the DC bus voltage, the motor phase current or the DC bus current, the ambient temperature, the switching frequency of the intelligent power module and the motor speed as inputs.

[0066] Specifically, when predicting the junction temperature of the intelligent power module, the junction temperature prediction model takes the DC bus voltage, the motor phase current or the DC bus current, the ambient temperature, the switching frequency of the intelligent power module and the motor speed as inputs, the input data can all affect the junction temperature of the intelligent power module, and inputting the related data can improve the accuracy of the junction temperature prediction value of the intelligent power module output by the junction temperature prediction model.

[0067] The fluctuation of the DC bus voltage will change the switching loss and the conduction loss of the power device. The switching loss is in a positive relationship with the square of the DC bus voltage, that is, the switching loss will increase sharply as the DC bus voltage increases. The conduction loss will also change with the change of the DC bus voltage, because the conduction resistance of the power device may change slightly at different voltages. The increase of these losses will eventually cause the junction temperature to rise.

[0068] The conduction loss is in a positive relationship with the square of the current, that is, if the current doubles, the conduction loss will increase by four times. In addition, a larger current will also make the switching process of the power device more difficult, and the switching time will become longer, thereby increasing the switching loss. The accumulation of these losses will cause the junction temperature to rise rapidly, and if the junction temperature is too high, it may exceed the rated temperature of the power device, causing the device to be damaged.

[0069] The ambient temperature is an important boundary condition in the junction temperature prediction model. Generally speaking, the junction temperature of the intelligent power module may increase by 10℃ for every 10℃ increase in ambient temperature, and the specific increase depends on the heat dissipation design and thermal resistance of the intelligent power module. In actual application, the range of ambient temperature changes needs to be considered to ensure that the intelligent power module can work safely and reliably under various environmental conditions.

[0070] Switching loss is directly proportional to switching frequency, that is, if the switching frequency doubles, the switching loss will also double. In addition, higher switching frequency will also generate more electromagnetic interference, which may require additional filtering and shielding of the circuit, but this will also increase the complexity and cost of the system. Therefore, when selecting the switching frequency, the control performance of the motor and the junction temperature of the intelligent power module need to be considered comprehensively.

[0071] The change of motor speed will indirectly affect the junction temperature by affecting the motor current, switching loss and conduction loss, etc. At high speed and light load, although the motor current is small, the switching loss may be large; at low speed and heavy load, conduction loss will be the main factor. Therefore, in the junction temperature prediction model, the comprehensive influence of the change of motor speed on various losses needs to be considered.

[0072] The five input parameters, i.e. the DC bus voltage, the motor phase current or the DC bus current, the ambient temperature, the switching frequency of the intelligent power module and the motor speed, affect the heating of the intelligent power module from different aspects. The junction temperature prediction model can accurately predict the junction temperature value of the intelligent power module by comprehensively considering these parameters.

[0073] In some embodiments, the DC bus voltage is a voltage signal subjected to preset voltage amplitude fluctuation processing on the original DC bus voltage.

[0074] Specifically, in the junction temperature prediction model, the original DC bus voltage is subjected to preset voltage amplitude fluctuation processing, which is usually to simulate the instability of the DC bus voltage under actual working conditions. In actual operation, the DC bus voltage will be affected by power grid fluctuations, power quality, load changes and other factors, resulting in fluctuations in its amplitude within a certain range. By presetting the fluctuation processing, the junction temperature prediction model can be more close to the actual running environment, and the accuracy and reliability of the junction temperature prediction model can be improved; in order to make the characteristics of the input parameters of the junction temperature prediction model meet the actual use scene better, the collected input signals need to be preprocessed, for example, adding ±20% random disturbance to the DC bus voltage signal to simulate the fluctuation of the power grid

[0075] For example, the current industry air conditioner outdoor unit controller temperature monitoring mode of intelligent power module is mainly divided into two kinds: hardware scheme is to detect the temperature of the heat sink through the temperature sensor, or to output the temperature signal collected through the temperature sensor of the intelligent power module itself, this kind of way exists the delay of heat transfer; Software scheme is to build a linear model to estimate the junction temperature by combining some electrical parameters with thermal resistance formula, this kind of way assumes that the current and temperature rise are linearly related, and does not consider the influence of power grid distortion.

[0076] The existing scheme has the following problems: the temperature sensor method cannot directly collect the module junction temperature, the heat conduction path is long, there is a certain response delay, the temperature transient cannot be captured, and the temperature rise changes relatively lag. If the input voltage of the power grid exists abnormal fluctuation, for example, when the alternating input voltage increases, the power loss of the intelligent power module will increase, the heat is serious, and the temperature collection value will be delayed for a long time to reflect the change of the temperature, which deviates greatly from the actual working junction temperature, and effective protection cannot be realized; When the fixing condition of the heat sink is not ideal, the heat transfer effect will be worse, the working temperature of the power device in the module may remain at a high level for a long time, and timely intervention and protection cannot be performed, and there is a risk of damage; The linear estimation model cannot model the nonlinear change of switch loss caused by voltage mutation and voltage harmonic, and the error increases nonlinearly with the load rate.

[0077] The present application specifically relates to a kind of intelligent power module junction temperature real-time prediction and active protection method based on junction temperature prediction model, especially suitable for the intelligent power module overheat early warning and protection of air conditioner outdoor unit electric control system under the scene of abnormal fluctuation of power grid voltage.

[0078] The present application proposes a kind of prediction mode of intelligent power module junction temperature, by building a kind of fusion multi-source time series data junction temperature prediction model, can accurately identify abnormal condition, earlier than the sensor temperature signal collected by intelligent power module to identify the abnormal change of module junction temperature, by hierarchical protection and early take different measures to intervene the working state of intelligent power module, to avoid the damage of module.

[0079] The junction temperature prediction model can realize the prediction of junction temperature peak value in advance by several seconds through the cooperative mechanism of multi-dimensional time series data correlation modeling and dynamic threshold protection strategy, improve the protection response speed; Solve the problem that the junction temperature of intelligent power module cannot be directly measured in the actual use of air conditioner, break through the dependence of single sensor; At the same time, through the form of prediction, early identification and early intervention can be realized, a software and hardware cooperative protection mechanism is constructed, the intelligent power module failure rate is reduced, the machine downtime protection frequency is reduced, the maintenance cost is reduced, and the user experience is improved.

[0080] In some embodiments, the controller is further configured to obtain the junction temperature dynamic threshold value by a junction temperature dynamic threshold model, wherein the junction temperature dynamic threshold model takes the measured ambient temperature and the motor current as inputs.

[0081] Specifically, the increase of the ambient temperature can increase the junction temperature of the intelligent power module, and the motor current reflects the actual working current of the device over a period of time. In the present application, the motor current can be the current of the fan motor of the outdoor unit of the air conditioner or the current of the compressor motor of the outdoor unit of the air conditioner. The larger the motor current value is, the greater the heat generated inside the device is, and the higher the possibility of temperature rise is.

[0082] The junction temperature dynamic threshold model takes the measured ambient temperature and the motor current as inputs, dynamically adjusts the junction temperature threshold value, and provides an important guarantee for the safe operation of the intelligent power module. The junction temperature dynamic threshold model can obtain the junction temperature dynamic threshold value of the intelligent power module according to the actual working state and environmental conditions of the intelligent power module.

[0083] In some embodiments, the junction temperature dynamic threshold model is represented as follows:

[0084] Tj_threshold = Tj_base + K1×(Tamb-25) + K2×(Iavg / Irated);

[0085] wherein Tj_threshold is the junction temperature dynamic threshold value, Tj_base is the temperature safety threshold value of the intelligent power module, Tj_base = 125℃, K1 is the ambient compensation coefficient, Tamb is the measured ambient temperature, K2 is the motor load rate compensation coefficient, Iavg is the average motor current, and Irated is the rated motor current.

[0086] Specifically, after obtaining the junction temperature prediction value Tj_pre of the intelligent power module by the junction temperature prediction model, it is necessary to compare it with the junction temperature dynamic threshold value Tj_threshold to determine whether the temperature of the intelligent power module is too high, and then determine whether certain protection measures need to be taken. A single determination standard is easy to cause misjudgment, and therefore the setting of the junction temperature dynamic threshold value can improve the accuracy of the determination. The above data are obtained in real time as inputs of the junction temperature dynamic threshold model, so as to obtain the junction temperature dynamic threshold value of the intelligent power module under different states.

[0087] In some embodiments, the controller implements a hardware-software cooperative protection strategy and is configured to perform at least one of the following when executing the target protection strategy for the intelligent power module according to the temperature difference condition:

[0088] When the temperature difference condition is Tj_pre ≥ Tj_threshold +5℃, the target protection strategy is: performing fault alarm, immediately shutting down the control signal of the intelligent power module, wherein Tj_pre is a junction temperature prediction value, and Tj_threshold is a junction temperature dynamic threshold;

[0089] When the temperature difference condition is Tj_threshold ≤Tj_pre<Tj_threshold +5℃, the target protection strategy is: performing fault alarm, the intelligent power module is reduced to 50% and forced cooling is started;

[0090] When the temperature difference condition is Tj_threshold -10℃≤Tj_pre<Tj_threshold, the target protection strategy is: triggering the PWM duty cycle linear reduction of the intelligent power module control signal;

[0091] When the temperature difference condition is Tj_pre<Tj_threshold -10℃, the target protection strategy is: allowing the intelligent power module to run in an overclocking mode (the highest +10% switching frequency).

[0092] Specifically, after obtaining the junction temperature prediction value of the intelligent power module through the junction temperature prediction model, the junction temperature prediction value is compared with the junction temperature dynamic threshold to determine whether the temperature of the intelligent power module is high, and then whether certain protection measures need to be taken is determined. When the intelligent power module is protected and controlled, the hierarchical protection strategy is determined. By comparing the predicted junction temperature with the junction temperature dynamic threshold, a temperature difference condition can be obtained, which is used to reflect the degree of failure. Different protection actions can be set according to the temperature difference condition, and then efficient protection is realized. The specific protection actions are shown in Table 1:

[0093] Table 1: Intelligent power module protection strategy table

[0094]

[0095] For example, based on the establishment of the junction temperature prediction model, the junction temperature of the intelligent power module can be predicted. The DC bus voltage, motor phase current or DC bus current, ambient temperature, switching frequency of the intelligent power module, motor speed, etc. are set as input parameters, and the intelligent power module junction temperature prediction value is set as an output parameter. A large number of input parameters and output parameters are collected under test conditions, and the intelligent power module junction temperature needs to be collected by the internal pre-embedded thermocouple. The collected parameters are used as training data for the junction temperature prediction model. The junction temperature prediction model can obtain the non-linear relationship between the input parameters and the output parameters. By adjusting the network parameters, a good training effect can be obtained. In the use process of the junction temperature prediction model, the input parameters in the last time step can be used to predict the output parameters at the next time, so that the junction temperature of the intelligent power module can be predicted by directly collecting the electrical parameters and the ambient temperature parameters. Then, by comparing with the junction temperature dynamic threshold, the grading protection is realized. Different processing measures are taken according to the degree of temperature rise abnormality, the intelligent power module high temperature early warning and protection are realized efficiently and quickly, and the abnormal damage of the machine is avoided.

[0096] The present application proposes a method for predicting the junction temperature of an intelligent power module based on a junction temperature prediction model, which is used for over-temperature protection of the intelligent power module. The present application constructs a technical chain of "data collection->junction temperature prediction model modeling->junction temperature dynamic threshold determination->grading protection", realizes the early prediction and active intervention of the junction temperature of the intelligent power module through the multi-source time sequence fusion of voltage, current, environmental parameters and the like; as shown in Figure 6 The corresponding data is obtained, preprocessed and input into the junction temperature prediction model, and finally the junction temperature prediction value of the intelligent power module is output by the junction temperature prediction model; while the junction temperature prediction model is predicting the junction temperature of the intelligent power module, the junction temperature dynamic threshold model obtains the junction temperature dynamic threshold of the intelligent power module according to the motor current and the ambient temperature compensation, and stores it in the dynamic threshold library. The temperature difference condition is obtained by comparing the junction temperature prediction value with the junction temperature dynamic threshold, and the intelligent power module is protected in stages according to the temperature difference condition.

[0097] In some embodiments, the junction temperature prediction model is a recurrent neural network model with a forgetting gate mechanism.

[0098] Specifically, the junction temperature prediction model is a recurrent neural network with a forgetting gate mechanism. By introducing forgetting gate, input gate and output gate mechanisms, the gradient vanishing and gradient explosion problems are effectively solved, the learning ability of the model for long-term dependence is improved, and the forgetting gate mechanism enables the junction temperature prediction model to selectively discard and retain information, thereby accurately predicting the junction temperature peak of the intelligent power module.

[0099] In some embodiments, the recurrent neural network model with a forgetting gate mechanism includes one of a long short-term memory network model and a network model derived based on the long short-term memory network model.

[0100] Specifically, the network model derived based on the long short-term memory network model includes: a gated recurrent unit network model, a bidirectional long short-term memory network model, a long short-term memory network student model obtained by compressing the long short-term memory network model through a knowledge distillation technology, etc.; or a network model formed by combining the long short-term memory network model with other strategies, such as a model combining the long short-term memory network with an attention mechanism or a model combining the long short-term memory network with a transfer learning technology.

[0101] The long short-term memory network model is a special recurrent neural network that controls the flow of information by introducing a gating mechanism (input gate, forget gate, and output gate). The junction temperature prediction usually needs to consider the historical running data of the device, such as temperature, current, voltage, etc. in a period of time, and the long short-term memory network model can effectively process these long sequence data and capture the long-term dependencies therein. For example, during the long-time running of the device, the temperature change in the early stage may have an impact on the current junction temperature, and the long short-term memory network model can remember these important historical information through its gating mechanism, thereby improving the accuracy of the prediction. When pre-training the long short-term memory network model, the historical data of intelligent power modules of different brands and different power levels are introduced to enhance the generalization.

[0102] The gated recurrent unit network model is a variant of the long short-term memory network model, which combines the input gate and the forget gate of the long short-term memory network model into an update gate and introduces a reset gate. The update gate is similar to the combination of the forget gate and the input gate in the long short-term memory network model, which determines how much information at the current time needs to be updated to the new hidden state. The greater the output value of the update gate, the more historical information is retained and the less new information is updated. The reset gate is used to control the influence of the hidden state of the previous time on the current candidate hidden state. If the output value of the reset gate is close to 0, the influence of the hidden state of the previous time on the current candidate hidden state will be ignored; if it is close to 1, the hidden state of the previous time will fully participate in the calculation of the current candidate hidden state.

[0103] The structure of the gated recurrent unit network model is simpler and has fewer parameters than the long short-term memory network model, so it is more computationally efficient during training and inference. When predicting junction temperature on resource-constrained devices, the gated recurrent unit network model may be a better choice. Although the structure of the gated recurrent unit network model is simplified, its performance is comparable to that of the long short-term memory network model in many tasks. In junction temperature prediction, if the dataset is not particularly large or there are strict limitations on computing resources, the gated recurrent unit network model can reduce computation time and resource consumption while ensuring a certain level of prediction accuracy. If the long short-term memory network model is insufficient in computing resources, a lightweight gated recurrent unit network model can be used instead, sacrificing some accuracy to reduce computational load.

[0104] The student model of the long short-term memory network is compressed through the knowledge distillation technique, reducing the number of model parameters and computational complexity, and reducing the model size by 75%, achieving model lightweight. This allows the model to be deployed on resource-constrained devices such as embedded systems or mobile devices while maintaining good prediction performance. Because the student model has a simpler structure, it is faster in inference. In real-time junction temperature prediction applications, fast inference can quickly obtain device junction temperature information to take appropriate control measures such as adjusting the cooling strategy or reducing device load.

[0105] The bidirectional long short-term memory network model is a variant of the long short-term memory network model, composed of a forward long short-term memory network and a backward long short-term memory network. The forward long short-term memory network processes sequence data in chronological order, while the backward long short-term memory network processes sequence data in reverse chronological order. The hidden states of the two networks are finally concatenated to obtain a more comprehensive sequence representation, and the bidirectional long short-term memory network model can capture more context information before and after the sequence data.

[0106] Attention mechanisms allow the model to dynamically focus on different parts of the sequence when processing sequence data. Attention mechanisms assign a weight to each element in the sequence, with a larger weight indicating that the element is more important to the current task. By incorporating attention mechanisms, the long short-term memory network model can more accurately capture key information in the sequence. Transfer learning techniques can transfer knowledge learned in one task or domain to another related task, reducing the dependence on large amounts of labeled data. The model combining long short-term memory networks and attention mechanisms can improve sequence data processing accuracy, and the model combining long short-term memory networks and transfer learning can use source domain knowledge to accelerate target task learning. By designing a reasonable model architecture and training strategy, the advantages of both techniques can be fully utilized.

[0107] For example, the structure of the long short-term memory network model is as shown in Figure 7 , Xt andht are the input and output of time step t, respectively. The neural network module A takes the input Xt and outputs ht the value of. However, the output of time t is not only related to the input of the current time point, but also related to the information of previous time points. Such a loop allows information to be continuously passed from the current step to the next step.

[0108] The long short-term memory network model has a design structure called "gate" that can realize the ability to remove or add information in the cell state. The "gate" is a component that allows information to pass selectively, which contains a sigmoid function and a multiplication operation. The principle of a normal long short-term memory network unit is shown in Figure 8 A long short-term memory network unit has an input gate, a forget gate, and an output gate to protect and control the input information. The input gate can select which information can pass, the forget gate can selectively ignore some information of the current stage, and the output gate finally decides which information can be output as the current state and updates the hidden state ht-1 The formula of long short-term memory network is as follows:

[0109] t = tanh(W xc x t +W hc h t-1 +b c ) ;

[0110] i t = sigm (W xi x t +W hi h t-1 +b i ) ;

[0111] f t = sigm (W xf x t +W hf h t-1 +b f ) ;

[0112] O t = sigm (W xo x t +W ho h t-1 +b o ) ;

[0113] C t =f t C t-1 +i t t ;

[0114] h t =O t tanh(C t ) ;

[0115] it, ft and Ot represent the input gate, the forget gate and the output gate, respectively, Xt is the input at the current time, ht is the output at the current time, ht-1 is the hidden state at the previous time, Ct and Ct-1 are storage units, t represents the storage state at the current time, b is a bias vector, which can make the network more flexible. The weights W represent the conversion between the two parts, for example, W xf represents the weight matrix of the input-forget gate, W hf represents the weight matrix of the hidden-forget gate.

[0116] The forget gate determines which information in the cell state needs to be discarded when the LSTM unit is working. The forget gate takes the hidden state at the previous time ht-1 and the input at the current time xt, splices ht-1 and xt into a vector, which is converted through the weight matrix W xf and the bias vectorb f Linear transformation is performed and the output is compressed to 0 to 1 by a Sigmoid function.

[0117] The input gate is multiplied by a weight matrix W xi and a bias vector b i Linear transformation is performed on the concatenated vector and the output is compressed to 0 to 1 by a Sigmoid function.

[0118] The storage state at the current time t The forget gate is multiplied by a weight matrix W xc and a bias vector b c Linear transformation is performed on the concatenated vector and the output is compressed to -1 to 1 by a tanh function.

[0119] The storage state Ct is updated according to the outputs of the forget gate and the input gate, the forget gate ft is multiplied element-wise by the storage state at the previous time Ct-1 , discarding information that needs to be discarded, and the input gate it is multiplied element-wise by the storage state at the current time t is multiplied element-wise by the storage state at the previous time

[0120] Ot The output gate determines the output state at the current time, and is multiplied element-wise by a weight matrix W xo and a bias vector b o Linear transformation is performed on the concatenated vector and the output is compressed to 0 to 1 by a Sigmoid function.

[0121] The output gate Ot is multiplied element-wise by the activated storage state Ct at the current time, and the output state at the current time ht is obtained.

[0122] sigm is a sigmoid activation function, and tanh is a hyperbolic function, which is defined as:

[0123] ;

[0124] ;

[0125] F represents the corresponding function function, the sigm function outputs a value from 0 to 1 as the gating state. The tanh function converts the result into a value from -1 to 1, which is used as the input signal.

[0126] The training effect of the long short-term memory network mainly depends on the training data and the adjustment of the network parameters. The training data mainly includes two dimensions: one is the amount of training data, and the amount of data collected under the test conditions is as large as possible; the other is the quality of the training data, and the collected data should be as diverse as possible, considering different device states and different working conditions and other actual situations; under the condition that the amount and quality of the training data are sufficient, the error of the verification data is used to judge the training effect of the network, and the parameters of the network, especially the number of hidden layers, the number of nodes, the batch size and the time step, are adjusted until the network has good prediction effect.

[0127] In some embodiments, the long short-term memory network model includes an input layer, a hidden layer and an output layer, wherein the input layer is configured to input parameters of a preset time step, the hidden layer includes a double-layer long short-term memory network, the first layer long short-term memory network is used to extract time sequence features, the second layer long short-term memory network is used to capture long-period cumulative effect, a preset Dropout value is added between the first layer long short-term memory network and the second layer long short-term memory network, and the output layer is used to output the junction temperature prediction value of the intelligent power module; the loss function of the long short-term memory network model adopts Huber loss, and the optimizer of the long short-term memory network model adopts Adam algorithm.

[0128] Specifically, the long short-term memory network model is usually composed of an input layer, a hidden layer and an output layer, and this hierarchical structure enables it to effectively process sequence data. The input layer is configured to input parameters of a preset time step, and the time step refers to the number of consecutive data points considered by the model at a time when processing sequence data. By setting an appropriate time step, the input layer can convert sequence data into a format suitable for processing by the long short-term memory network model. Different time steps will affect the model's ability to capture historical information. A shorter time step may not be able to fully utilize sufficient historical information, resulting in reduced prediction accuracy; while a longer time step, although it can include more historical information, will increase the computational complexity and training difficulty of the model.

[0129] The hidden layer includes a double-layer long short-term memory network, and each layer of long short-term memory network is composed of multiple long short-term memory network units. These units control the flow of information through a gating mechanism (input gate, forget gate and output gate).

[0130] The first layer of long short-term memory network is used to extract time series features, responsible for receiving data from the input layer and preliminarily extracting features in the sequence data; the second layer of long short-term memory network is used to capture long-period cumulative effect, receiving the output of the first layer of long short-term memory network as input, further abstracting and extracting features, spanning longer time steps, and associating local features at different time points to discover long-period cumulative effect. Through the stacking of double-layer long short-term memory network, the long short-term memory network model can learn deeper feature representation in the data, thereby improving the accuracy of prediction.

[0131] In junction temperature prediction, the output layer usually outputs a predicted junction temperature value, which can be predicted according to the output of the hidden layer. For example, if the current of the intelligent power module continues to increase in the past period of time, the ambient temperature also rises, and the long short-term memory network model may predict that the junction temperature will rise accordingly.

[0132] Dropout is a regularization technique. During the training of neural networks, it randomly sets the output of a portion of neurons to zero according to a predetermined probability value (i.e., a predetermined Dropout value). These zeroed neurons do not participate in the forward propagation and back propagation processes in this training iteration, which is equivalent to using a "simplified version" of the neural network in each iteration. In junction temperature prediction, if the long short-term memory network model performs well on the training data but the performance decreases on the test data, it indicates that the model has overfitting phenomenon. Adding Dropout between the double-layer long short-term memory network can prevent the model from over-relying on certain specific neuron connections. Dropout randomly discards different neurons in each iteration, which is equivalent to training multiple different sub-models. These sub-models can be regarded as an integration of the overall model when predicting, and by integrating the prediction results of multiple sub-models, the risk of overfitting of a single model can be reduced, and the stability and accuracy of the overall model can be improved.

[0133] Huber loss is a loss function that combines the advantages of mean square error and mean absolute error. The long short-term memory network model may have some abnormal data points in the sequence data, such as sudden failure of the device or temporary error of the sensor. Huber loss function can effectively handle these abnormal data, ensuring that the model will not deviate from the correct prediction direction in the training process due to individual outliers.

[0134] Adam (Adaptive Moment Estimation) algorithm is an adaptive learning rate optimization algorithm. Adam algorithm can adaptively adjust the learning rate according to the historical gradient information of each parameter. For parameters with large gradient changes, the learning rate will automatically decrease to avoid excessive parameter updates. For parameters with small gradient changes, the learning rate will automatically increase to speed up the parameter update speed. This makes the Adam algorithm more efficient in finding the optimal solution for complex neural networks such as long short-term memory network models.

[0135] For example, when predicting the junction temperature of an intelligent power module based on a long short-term memory network, data collection and preprocessing are first performed. The heat of the intelligent power module is mainly caused by the working loss of the internal power device. The junction temperature of the intelligent power module is related to multiple variables such as electrical parameters, environmental parameters, and operating parameters. During the operation of the air conditioner, the parameters that can be normally collected by the whole machine are selected as input parameters, and the module junction temperature that cannot be directly collected is selected as an output parameter. The types of input parameters are as follows:

[0136] Electrical parameters: DC bus voltage Vdc, motor phase current Iu / Iv / Iw, or DC bus current Idc;

[0137] Environmental parameters: ambient temperature Tamb (NTC sensor);

[0138] Operating parameters: IPM switching frequency fsw, motor speed Rfan.

[0139] The collection frequency of input parameters is set to 1 kHz, and the sliding window method is used to generate training samples. The window length is set to 10 seconds, and the step size is 0.1. At the same time, in order to make the features of the input parameters more suitable for the actual use scenario, the collected input signals need to be preprocessed, such as adding ±20% random disturbance to the voltage signal to simulate the fluctuation of the power grid.

[0140] After data collection and preprocessing, the long short-term memory network model is built:

[0141] The basic long short-term memory network mainly includes three network structures: input layer, hidden layer, and output layer.

[0142] The input layer adopts seven-dimensional feature input or five-dimensional feature input, each feature adopts 100 sampling points, that is, 100 sampling points * 7-dimensional features (Vdc, Iu, Iv, Iw, Tamb, fsw, Rfan) or 5-dimensional features (Vdc, Idc, Tamb, fsw, Rfan), wherein Vdc is the DC bus voltage, Iu is the motor U-phase current, Iv is the motor V-phase current, Iw is the motor W-phase current, Idc is the motor DC bus current, Tamb is the ambient temperature, fsw is the switching frequency of the intelligent power module, and Rfan is the motor speed.

[0143] The hidden layer adopts a double-layer long short-term memory network, the first layer of the long short-term memory network is provided with 64 units and is used for extracting time sequence features, the second layer of the long short-term memory network is provided with 32 units and is used for capturing long-period cumulative effects, such as the influence of slow rising of the ambient temperature on the junction temperature, the activation function is Tanh, Tanh can better process positive and negative features in the long short-term memory network, such as the up and down fluctuations of the voltage, to avoid the disappearance of the gradient, a Dropout rate of 0.2 is added between the first layer and the second layer to randomly shield 20% of the nodes to prevent the model from overfitting to a single data and excessively relying on a certain parameter while ignoring the influence of other parameters.

[0144] The output layer is used for outputting the junction temperature peak value of the intelligent power module within the next 3 seconds.

[0145] The loss function of the long short-term memory network model adopts Huber loss, and the optimizer adopts the Adam algorithm.

[0146] The Huber loss (δ=1.0) is adopted to balance the sensitivity of MSE (mean square error) to abnormal values. In the case of δ=1.0, the Huber loss adopts the form of square error when the prediction error is less than or equal to 1.0, and adopts the form of mean absolute error when the prediction error is greater than 1.0. This setting enables the long short-term memory network model to fully utilize the smoothness of the mean square error when the error is small, and quickly converges; when the error is large, it can reduce the influence of abnormal values and improve the robustness of the long short-term memory network model.

[0147] The optimizer adopts the Adam (Adaptive Moment Estimation) algorithm to adjust appropriate initial learning rate and batch size parameters. The Adam algorithm provides strong optimization capability for long short-term memory network model training through its adaptive learning rate and momentum mechanism.

[0148] In some embodiments, the junction temperature prediction value of the intelligent power module is the junction temperature peak value of the intelligent power module within a preset time length in the future.

[0149] Specifically, the junction temperature of the intelligent power module is an important factor affecting the reliability and service life of the module, and high junction temperature can cause performance degradation, service life shortening, and even failure of the device. Therefore, it is necessary to accurately predict the junction temperature peak of the intelligent power module within a preset time in the future. Combining the predicted junction temperature peak prediction with an intelligent control algorithm can adjust the working state of the module in real time according to the prediction result, realize dynamic power distribution and thermal management, and improve the efficiency and intelligent level of the intelligent power module.

[0150] In the description of the present specification, the description referring to the terms "one embodiment", "some embodiments", "exemplary embodiment", "example", "specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the exemplary description of the above terms does not necessarily mean the same embodiment or example.

[0151] Although the embodiments of the present application have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the claims and their equivalents.

Claims

1. An air conditioner, comprising: A motor and a drive module, wherein the drive module is used to drive the motor to operate, and the drive module includes at least one intelligent power module; The voltage detection module is used to detect the DC bus voltage of the DC bus to which the drive module is connected; The current detection module is used to detect the phase current of the motor or the DC bus current. Temperature detection module, used to detect ambient temperature; The controller is connected to the drive module, the voltage detection module, the current detection module, and the temperature detection module. The controller is configured with a junction temperature prediction model and is also configured to obtain the junction temperature dynamic threshold through a junction temperature dynamic threshold model, wherein the junction temperature dynamic threshold model takes the measured ambient temperature and the motor current as inputs. The junction temperature dynamic threshold model is expressed as follows: Tj_threshold = Tj_base + K1×(Tamb-25) + K2×(Iavg / Irated); Wherein, Tj_threshold is the junction temperature dynamic threshold, Tj_base is the temperature safety threshold of the intelligent power module, Tj_base=125℃, K1 is the environmental compensation coefficient, Tamb is the measured ambient temperature, K2 is the motor load rate compensation coefficient, Iavg is the average motor current, and Irated is the rated motor current. The controller is configured to: The junction temperature prediction value of the intelligent power module is obtained through the junction temperature prediction model, wherein the junction temperature prediction model takes at least the DC bus voltage, the motor phase current or DC bus current and the ambient temperature as inputs and the junction temperature prediction model outputs the junction temperature prediction value of the intelligent power module. The junction temperature dynamic threshold of the intelligent power module is generated based on the measured ambient temperature and motor current. The target protection strategy for the intelligent power module is executed based on the temperature difference between the predicted junction temperature and the dynamic junction temperature threshold.

2. The air conditioner according to claim 1, characterized in that, The junction temperature prediction model uses the DC bus voltage, the motor phase current or DC bus current, the ambient temperature, the switching frequency of the intelligent power module, and the motor speed as inputs.

3. The air conditioner according to claim 1 or 2, characterized in that, The DC bus voltage is a voltage signal that has undergone preset voltage amplitude fluctuation processing on the original DC bus voltage.

4. The air conditioner according to claim 1 or 2, characterized in that, The controller implements a hardware-software collaborative protection strategy and is configured to perform at least one of the following when executing a target protection strategy for the intelligent power module based on the temperature difference conditions: When the temperature difference condition is Tj_pre ≥ Tj_threshold +5℃, the target protection strategy is: to issue a fault alarm and immediately shut down the control signal of the intelligent power module, where Tj_pre is the predicted junction temperature value and Tj_threshold is the dynamic threshold of the junction temperature. When the temperature difference condition is Tj_threshold ≤ Tj_pre < Tj_threshold +5℃, the target protection strategy is: to issue a fault alarm, and the intelligent power module reduces its frequency to 50% and starts forced cooling. When the temperature difference condition is Tj_threshold -10℃≤Tj_pre < Tj_threshold, the target protection strategy is: to trigger a linear reduction in the PWM duty cycle of the intelligent power module control signal; When the temperature difference condition is Tj_pre < Tj_threshold -10℃, the target protection strategy is: allow the intelligent power module to operate at overclock (up to +10% of the switching frequency).

5. The air conditioner according to claim 1 or 2, characterized in that, The junction temperature prediction model is a recurrent neural network model with a forget gate mechanism.

6. The air conditioner according to claim 5, characterized in that, The recurrent neural network model with forget gate mechanism includes one of the long short-term memory network model and a network model derived from the long short-term memory network model.

7. The air conditioner according to claim 6, characterized in that, The Long Short-Term Memory (LSTM) network model includes an input layer, a hidden layer, and an output layer. The input layer is configured with input parameters of a preset time step. The hidden layer includes a two-layer LSM network. The first LSM network is used to extract temporal features, and the second LSM network is used to capture long-term cumulative effects. A preset Dropout value is added between the first LSM network and the second LSM network. The output layer is used to output the junction temperature prediction value of the intelligent power module. The loss function of the Long Short-Term Memory (LSTM) network model is Huber loss, and the optimizer of the LSM network model is the Adam algorithm.

8. The air conditioner according to claim 7, characterized in that, The predicted junction temperature value of the intelligent power module is the peak junction temperature of the intelligent power module within a preset time period in the future.

Images