Temperature control method of temperature controller

Abstract

The invention relates to a temperature control method of a temperature controller, which comprises the following steps: S1, collecting multi-source heterogeneous data, fusing the collected data through an adaptive weighted Kalman filtering algorithm, and outputting a temperature field model, S2, inputting the temperature field model into a double-layer LSTM neural network, calculating the importance weight of each region of the temperature field model, and outputting a real-time temperature field and a time derivative dT / dt thereof, s4, calculating a heat capacity parameter C and a heat resistance parameter R in real time, and adjusting a PID (Proportion Integration Differentiation) parameter through a fuzzy rule base; S5, switching to a pulse strong control mode when dT / dt is detected to be greater than or equal to 5 DEG C, and switching to a PID mode when delta T is less than or equal to 5 DEG C and is stable for 2 seconds; the system has the advantage of accurate control.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of vehicle temperature control, and particularly relates to a temperature control method of a temperature controller. BACKGROUND

[0002] The temperature controller, also known as a temperature control switch, a temperature protector, or a temperature controller, refers to a series of automatic control elements that can produce a certain special effect and produce on or off actions according to the temperature change of the working environment, through the physical deformation of the switch.

[0003] In the prior art, the temperature controller on the automobile can control the temperature based on the temperature in the vehicle, but due to the traditional single-point temperature measurement (such as a thermocouple) or low-resolution infrared imaging, the spatial temperature field modeling is distorted, so that the micro-area temperature gradient cannot be captured, and the control accuracy is affected. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide a temperature control method of a temperature controller with precise control.

[0005] The technical solution of the present application is as follows:

[0006] A temperature control method of a temperature controller, comprising the following steps

[0007] S1: Collecting multi-source heterogeneous data and fusing the collected data through an adaptive weighted Kalman filtering algorithm and outputting a temperature field model;

[0008] S2: Inputting the temperature field model into a double-layer LSTM neural network, calculating the importance weight of each region of the temperature field model, and outputting a real-time temperature field and its time derivative dT / dt;

[0009] S3: Focusing on the hot spot region based on the importance weight of each region of the real-time temperature field, predicting the temperature, and generating a PWM instruction;

[0010] S4: Real-time calculation of the heat capacity parameter C and the thermal resistance parameter R, and adjustment of the PID parameter through a fuzzy rule base;

[0011] S5: When detecting ||dT / dt||≥5℃, switching to a pulse strong control mode, and when ||ΔT||≤5℃ and stable for 2s, switching to a PID mode.

[0012] Further, the step S1 specifically comprises the following steps:

[0013] S1.1: Acquiring the spatial temperature distribution through an 8x8 infrared thermal imaging sensor array, acquiring the contact point temperature through an embedded NTC thermocouple matrix, and acquiring the actuator current ripple through a high-precision current sensor, so as to realize the acquisition of multi-source heterogeneous data;

[0014] S1.2: Spatio-temporal alignment is performed, taking thermocouple data as the reference, and adaptive weighted Kalman filtering is used to dynamically allocate the weight of infrared data, so as to realize the fusion of multi-source heterogeneous data;

[0015] S1.3: The fused data is mapped to a three-dimensional space grid with a resolution of 1 cm 3 , and a continuous temperature field model is generated by a bicubic spline interpolation algorithm.

[0016] Further, the step S1 further includes the following steps:

[0017] S1.4: 1024-point FFT transformation is performed on the current signal;

[0018] S1.5: When the amplitude of the 72±5Hz component is greater than 3 times the baseline value and the current fluctuation variance is greater than 0.5A 2 , it is determined that the compressor bearing wear fault occurs and the power reduction protection strategy is started;

[0019] S1.6: When the 150±5Hz component lasts for more than 5 seconds and the average current drops by 10%, it is determined that the refrigerant leakage fault occurs, triggering the redundant refrigeration compensation, and transferring the refrigeration load to the non-fault partition while increasing the fan speed by 20%.

[0020] Further, the step S2 specifically includes the following steps:

[0021] S2.1: The temperature field is divided into 8x8 partitions according to the layout of the infrared thermal imaging sensor array, and each partition corresponds to a single infrared thermal imaging sensor;

[0022] S2.2: The LSTM neural network calculates the partition weight through the Attention mechanism;

[0023] S2.3: The spatial temperature gradient features are extracted through the CNN convolution layer to identify micro mutations of <0.5℃, and the real-time temperature field and its time derivative dT / dt of the whole field are output at the same time.

[0024] Further, the step S3 specifically includes the following steps:

[0025] S3.1: The LSTM neural network automatically focuses on the partition with a temperature change rate greater than 2℃ / s;

[0026] S3.2: The temperature curve of the target partition in the next 30s is predicted, and the duty cycle command is generated according to the formula PWM 调整 =Kp×(T 预测 -T 设定 ), where Kp is the proportional coefficient;

[0027] S3.3: When the prediction error > 0.5 DEG C, switch to the Bayesian optimization algorithm to regenerate the instruction, and when the prediction error <= 0.5 DEG C, maintain the current PWM instruction unchanged.

[0028] Further, the step S4 specifically comprises the following steps:

[0029] S4.1: a step current with a pulse width of 10ms and an amplitude of 2A is applied to the actuator, and a temperature rise response curve is synchronously collected;

[0030] S4.2: a first-order model ΔT(t) = ΔQ·(1-e -t / (R·C) ) is fitted, and the heat capacity parameter C and the thermal resistance parameter R are inversely solved;

[0031] S4.3: a fuzzy rule base is constructed with temperature deviation |ΔT| and |dT / dt| as inputs, and rules are constructed to adjust PID parameters;

[0032] S4.4: the fuzzy rule threshold is dynamically contracted according to the overshoot through a steady-state time feedback module.

[0033] Further, it further comprises performing three-level superposition correction of the predicted temperature, and the three-level superposition correction of the predicted temperature specifically comprises the following steps:

[0034] S6.1: a fuzzy controller is called based on a vehicle speed signal, and a temperature is compensated according to a mapping table;

[0035] S6.2: a compensation amount is superimposed based on an air conditioner air outlet mode;

[0036] S6.3: a passenger's partition is compensated by ±0.5 DEG C based on a seat pressure sensor signal;

[0037] S6.4: when a human body surface temperature > 32 DEG C is detected, a-0.8 DEG C active cooling compensation is triggered;

[0038] S6.5: the execution priority is set as S6.4 > S6.3 > S6.2 > S6.1.

[0039] Compared with the prior art, the present application has the following beneficial effects:

[0040] 1、The present application adopts multi-source heterogeneous data to construct a temperature field model to eliminate single sensor error, and automatically identifies a hot spot area through an LSTM neural network, thereby improving resource utilization, and generates a PWM instruction based on a predicted temperature, reduces control shock, and when a sudden working condition is out of control, switches between a pulse strong control mode and a PID mode, thereby ensuring temperature control.

[0041] In summary, the present application has the advantages of precise control. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0043] A temperature control method of a temperature controller, comprising the following steps:

[0044] S1: collecting multi-source heterogeneous data and fusing the collected data through an adaptive weighted Kalman filtering algorithm and outputting a temperature field model;

[0045] S1.1: acquiring spatial temperature distribution through an 8*8 infrared thermal imaging sensor array (wavelength range 8-14 μm), acquiring contact point temperature through an embedded NTC thermocouple matrix (interval ≤ 15 mm), and acquiring actuator current ripple through a high-precision current sensor (sampling rate 10 kHz), so as to realize acquisition of multi-source heterogeneous data;

[0046] S1.2: performing space-time alignment, taking thermocouple data as a reference, and dynamically distributing infrared data weight through an adaptive weighted Kalman filtering, so as to realize fusion of multi-source heterogeneous data, and a weight formula is ω=1 / (σ 2 +δ), wherein σ is a standard deviation of sensor error, and δ is obtained through regression fitting of environmental temperature and humidity historical data, the formula is used for dynamically adjusting fusion weight according to sensor accuracy and environmental disturbance, and improving temperature field model reliability;

[0047] S1.3: mapping the fused data to a three-dimensional space grid with a resolution of 1 cm 3 , and generating a continuous temperature field model through a bicubic spline interpolation algorithm;

[0048] S1.4: performing 1024-point FFT transformation (Hanning window, frequency resolution 9.76 Hz) on the current signal, for compressor fault feature extraction;

[0049] S1.5: when a 72±5 Hz component amplitude > 3 times a baseline value and a current fluctuation variance > 0.5 A 2 , it is determined that the compressor bearing is in a wear fault state, and a power reduction protection strategy is started, wherein the power reduction protection strategy is to limit the compressor speed within a range of ±10% of the rated speed, and inject a harmonic current opposite to the 72 Hz abnormal vibration to offset the vibration, while reducing the upper limit of the PWM duty cycle to 70%;

[0050] S1.6: When the 150±5Hz component lasts for >5 seconds and the average current drops by 10%, determine a refrigerant leakage fault and trigger a redundant refrigeration compensation, wherein the redundant refrigeration compensation is that when a refrigerant leakage fault is determined, a standby electronic compressor refrigeration circuit (PTC heater auxiliary) is started, and the refrigeration load is transferred to a non-fault partition, while the fan speed is increased by 20% to enhance convective heat transfer;

[0051] S2: Input the temperature field model into a double-layer LSTM neural network, calculate the importance weight of each region of the temperature field model, and output the real-time temperature field and its time derivative dT / dt;

[0052] S2.1: Divide the temperature field into 8×8 partitions according to the layout of the infrared thermal imaging sensor array, and each partition corresponds to a single infrared thermal imaging sensor;

[0053] S2.2: The LSTM neural network calculates the partition weight (weight factor = real-time temperature change rate × historical abnormal number) through the Attention mechanism;

[0054] S2.3: Extract spatial temperature gradient features through the CNN convolution layer to identify micro mutations of <0.5℃, and simultaneously output the real-time temperature field and its time derivative dT / dt of the whole field;

[0055] S3: Focus on the hotspot area based on the importance weight of each region of the real-time temperature field, predict the temperature, and generate PWM instructions;

[0056] S3.1: The LSTM neural network automatically focuses on the partition with a temperature change rate >2℃ / s;

[0057] S3.2: Predict the temperature curve of the target partition in the next 30s, and generate the duty cycle instruction according to the formula PWM 调整 =K p ×(T 预测 -T 设定 ) (where K p is the proportional coefficient);

[0058] S3.3: When the prediction error is >0.5℃, switch to the Bayesian optimization algorithm to regenerate the instruction, and when the prediction error is ≤0.5℃, maintain the current PWM instruction unchanged;

[0059] S4: Real-time calculation of heat capacity parameter C and thermal resistance parameter R, and adjustment of PID parameters through fuzzy rule base;

[0060] S4.1: Apply a step current with a pulse width of 10ms and an amplitude of 2A to the actuator, and synchronously collect the temperature rise response curve;

[0061] S4.2: Fit a first-order model ΔT(t) = ΔQ·(1-e -t / (R·C)), inverse heat capacity parameter C and thermal resistance parameter R;

[0062] S4.3: Fuzzy rule base with |ΔT| (temperature deviation) and dT / dt (rate of change) as inputs to adjust PID parameters and build rules;

[0063] Where the rules are set as follows:

[0064] |ΔT|>3℃ and |dT / dt|>1℃ / s, then ΔKp=+0.2;

[0065] |ΔT|<1℃ and |dT / dt|<0.2℃ / s, then ΔKi=-0.05;

[0066] |ΔT|>2℃ and |dT / dt|<0.5℃ / s, then ΔKd=+0.15;

[0067] Where ΔKp is the proportional coefficient adjustment amount, ΔKi is the integral coefficient adjustment amount, and ΔKd is the differential coefficient adjustment amount.

[0068] Example rule: If |ΔT|>3℃ and dT / dt>1℃ / s, then the proportional coefficient Kp is increased by 0.2.

[0069] S4.4: Through the steady-state time feedback module, dynamically shrink the fuzzy rule threshold according to the overshoot (for example: when the overshoot is >10%, the threshold is shrunk to 80% of the original value), which is used to suppress PID parameter oscillation and ensure that the system converges quickly to steady state.

[0070] S5: When ||dT / dt||≥5℃ is detected, switch to pulse strong control mode, and when ||ΔT||≤5℃ and stable for 2s, switch to PID mode.

[0071] S6, three-level superimposed correction predicted temperature is performed, which specifically includes the following steps:

[0072] S6.1: Call the fuzzy controller based on the vehicle speed signal and compensate the temperature according to the mapping table (0-60km / h: 0℃, 60-120km / h: +0.3℃, >120km / h: +0.6℃);

[0073] S6.2: Superimpose compensation based on air conditioning air outlet mode (direct blowing mode: -0.4℃, diffusion mode: +0.2℃);

[0074] S6.3: Based on the seat pressure sensor signal, compensate ±0.5℃ for the passenger's zone;

[0075] S6.4: When the human body surface temperature is detected to be >32℃, trigger 0.8℃ active cooling compensation;

[0076] S6.5: The execution priority is set as S6.4 > S6.3 > S6.2 > S6.1.

[0077] Although the present application has been described in detail with reference to the foregoing embodiments, the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent features, by those skilled in the art, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A temperature control method for a thermostat, characterized in that: Includes the following steps S1: Collect multi-source heterogeneous data and fuse the collected data through adaptive weighted Kalman filter algorithm to output the temperature field model; S2: Input the temperature field model into the two-layer LSTM neural network, calculate the importance weight of each region of the temperature field model and output the real-time temperature field and its time derivative dT / dt; S3: Focuses on hotspots based on the importance weights of each area in the real-time temperature field, predicts the temperature and generates PWM instructions; S4: Calculate the thermal capacity parameter C and thermal resistance parameter R in real time, and adjust the PID parameters through the fuzzy rule base; S5: When it is detected that ||dT / dt||≥5℃, switch to pulse force control mode, and when ||ΔT||≤5℃ and remains stable for 2s, switch to PID mode.

2. The temperature control method of a thermostat according to claim 1, characterized in that: The step S1 specifically includes the following steps: S1.1: Acquire spatial temperature distribution using an 8×8 infrared thermal imaging sensor array, contact point temperature using an embedded NTC thermocouple matrix, and actuator current ripple using a high-precision current sensor to achieve multi-source heterogeneous data acquisition. S1.2: Perform spatiotemporal alignment, using thermocouple data as a benchmark and using adaptive weighted Kalman filtering to dynamically assign infrared data weights to achieve fusion of multi-source heterogeneous data; S1.3: Mapping fused data to 1cm 3 The continuous temperature field model is generated by the bicubic spline interpolation algorithm based on the three-dimensional space grid with high resolution.

3. The temperature control method of a thermostat according to claim 2, characterized in that: The step S1 further comprises the following steps: S1.4: Perform 1024-point FFT transformation on the current signal; S1.5: When the 72±5Hz component amplitude is detected to be >3 times the baseline value and the current fluctuation variance is >0.5A 2 When the fault occurs, it is determined to be a compressor bearing wear fault and the power reduction protection strategy is activated; S1.6: When the 150±5Hz component lasts for more than 5 seconds and the average current drops by 10%, it is determined to be a refrigerant leakage fault, triggering redundant cooling compensation, and transferring the cooling load to the non-faulty partition, while increasing the fan speed by 20%.

4. The temperature control method of a thermostat according to claim 3, characterized in that: The step S2 specifically includes the following steps: S2.1: Divide the temperature field into 8 × 8 zones according to the infrared thermal imaging sensor array layout, with each zone corresponding to a single infrared thermal imaging sensor; S2.2: LSTM neural network calculates partition weights through the Attention mechanism; S2.3: The spatial temperature gradient features are extracted through the CNN convolutional layer to identify micro-mutations <0.5°C and simultaneously output the real-time temperature field of the entire field and its time derivative dT / dt.

5. The temperature control method of a thermostat according to claim 4, characterized in that: The step S3 specifically includes the following steps: S3.1: LSTM neural network automatically focuses on the partition with temperature change rate > 2°C / s; S3.2: Predict the temperature curve of the target partition in the next 30 seconds, according to the PWM formula 调整 =K p ×(T 预测 -T 设定 ) generates a duty cycle instruction, where K p is the proportionality coefficient; S3.3: When the prediction error is greater than 0.5°C, switch to the Bayesian optimization algorithm to regenerate instructions. When the prediction error is less than or equal to 0.5°C, maintain the current PWM instruction unchanged.

6. The temperature control method of a thermostat according to claim 5, characterized in that: The step S4 specifically includes the following steps: S4.1: Apply a 10ms pulse width, 2A amplitude step current to the actuator and simultaneously collect the temperature rise response curve; S4.2: Fitting the first-order model ΔT(t) = ΔQ (1-e -t / (R·C) ), inversely solve the heat capacity parameter C and thermal resistance parameter R; S4.3: The fuzzy rule base takes the temperature deviation |ΔT| and |dT / dt| as input and constructs rules to adjust the PID parameters; S4.4: Dynamically shrink the fuzzy rule threshold according to the overshoot through the steady-state time feedback module.

7. The temperature control method of a thermostat according to claim 1, characterized in that: It also includes performing three-level superposition correction of predicted temperature, which specifically includes the following steps: S6.1: Call the fuzzy controller based on the vehicle speed signal and compensate the temperature according to the mapping table; S6.2: Compensation based on the air conditioner air outlet mode; S6.3: Based on the seat pressure sensor signal, compensate the passenger zone by ±0.5°C; S6.4: When the human body surface temperature is detected to be greater than 32°C, an active cooling compensation of -0.8°C is triggered; S6.5: The execution priority is set to S6.4>S6.3>S6.2>S6.1.