Search

CN-121976381-A - Multi-temperature-zone energy-saving control method based on reinforcement learning

CN121976381ACN 121976381 ACN121976381 ACN 121976381ACN-121976381-A

Abstract

The invention discloses a multi-temperature-zone energy-saving control method based on reinforcement learning, which relates to the technical field of automatic control and comprises four steps of clothes state sensing and modeling, multi-temperature-zone self-adaptive control, process optimization and learning, later evaluation and learning, wherein weight, moisture distribution, surface temperature and material data are collected through a plurality of sensors, moisture proportion is calculated, a thermodynamic diagram is drawn, a high-humidity zone is identified, a plurality of independent temperature zones are divided, temperature gradients are set on the basis of the material and the moisture distribution, power is dynamically adjusted, energy consumption is monitored in real time, an energy efficiency ratio optimization strategy is calculated, and the performance is improved through moisture proportion, weight change rate and safe temperature judgment through comprehensive scoring, reinforcement learning and knowledge base updating. The intelligent control system realizes accurate regulation and control of the drying process, has self-learning capability, improves energy utilization efficiency, and brings remarkable economic and environmental benefits to users on the premise of ensuring drying quality.

Inventors

  • CHEN LIXIN

Assignees

  • 台州市英络克工具有限公司

Dates

Publication Date
20260505
Application Date
20260130

Claims (10)

  1. 1. The multi-temperature-zone energy-saving control method based on reinforcement learning is characterized by comprising the following steps of: S1, acquiring initial weight and current weight of clothes through a high-precision weighing sensor, and calculating to obtain integral moisture proportion; Acquiring the moisture content distribution of different areas of the clothes through a microwave moisture sensor array, drawing a two-dimensional thermodynamic diagram to form a moisture distribution map, calculating the average value of the moisture of the areas, and marking a high-humidity area with the moisture content higher than the average value; acquiring the temperature distribution of the surface of the clothes by using a thermal infrared imager, and identifying the type of the clothes material by using near infrared spectrum analysis; integrating the data to construct a controlled object state model comprising the whole moisture proportion, the moisture distribution map, the high humidity area and the material type; S2, dividing the drying cabin into a plurality of independent temperature areas by a multi-temperature-area self-adaptive control strategy, determining basic temperature gradients of the temperature areas through preset rules based on the types of clothes materials, calculating temperature deviations of the temperature areas according to a moisture distribution map and a high-humidity area to adjust initial temperature parameters, calculating moisture change rate according to an average value of the moisture of the area, and self-adaptively controlling power of the temperature areas; S3, optimizing and learning a drying process, monitoring energy consumption of each temperature area in real time, calculating an overall energy efficiency ratio, dynamically adjusting a drying strategy according to the energy efficiency ratio, and judging whether drying is finished or not based on preset main judging conditions, auxiliary judging conditions and safety judging conditions; and S4, post-evaluation and system learning, namely comprehensively detecting the final humidity distribution of the clothes, calculating uniformity indexes, total energy consumption and total time, obtaining comprehensive scores, performing parameter self-optimization according to the comprehensive scores, and updating a knowledge base.
  2. 2. The reinforcement learning-based multi-temperature-zone energy-saving control method according to claim 1, wherein the multi-sensor comprises a high-precision weighing sensor, a microwave moisture sensor array, a thermal infrared imager and a near infrared spectrum analyzer; wherein, the high-precision weighing sensor is used for collecting initial weight And the current weight The microwave moisture sensor array is used for collecting moisture content distribution of different areas The thermal infrared imager is used for collecting the temperature distribution of the surface of the clothes The near infrared spectrum analyzer is used for identifying the material type of the clothes.
  3. 3. The reinforcement learning-based multi-temperature zone energy saving control method of claim 1, wherein the overall moisture ratio is calculated by the formula: And calculating, wherein the water distribution map is a two-dimensional water distribution thermodynamic diagram, and the high humidity area is an area with water content higher than an average value.
  4. 4. The reinforcement learning-based multi-temperature zone energy saving control method of claim 1, wherein the base temperature gradient is formulated as: The device is provided with a plurality of channels, Wherein N is the number of the temperature zones, Is the base temperature of the nth temperature zone, At the minimum of the base temperature of the furnace, Is the highest base temperature.
  5. 5. The reinforcement learning-based multi-temperature-zone energy saving control method according to claim 1, wherein the temperature zone parameters are expressed as: The adjustment is carried out so that the adjustment is carried out, Wherein the method comprises the steps of For the temperature adjustment value of the nth temperature zone, k is a proportionality coefficient, For the average moisture content of the nth temperature zone, Is the overall average moisture content.
  6. 6. The reinforcement learning-based multi-temperature-zone energy saving control method of claim 1, wherein the temperature zone power is as follows: The adjustment is carried out so that the adjustment is carried out, Wherein the method comprises the steps of For the adjusted nth temperature zone power, For the initial power to be applied, In order to adjust the coefficient of the power supply, The moisture change rate in the nth temperature range.
  7. 7. The reinforcement learning-based multi-temperature-zone energy-saving control method according to claim 1, wherein the energy consumption of each temperature zone is as follows: The calculation is performed such that, Wherein the method comprises the steps of For the energy consumption of the nth temperature zone, Real-time power for the nth temperature zone; The overall energy efficiency ratio is as follows: The calculation is performed such that, Wherein the method comprises the steps of In order to obtain the evaporation amount of water, Is the total energy consumption.
  8. 8. The reinforcement learning-based multi-temperature zone energy saving control method according to claim 1, wherein the main determination condition is: The auxiliary judgment condition is that The safety judgment condition is Wherein, the method comprises the steps of, As the threshold value of the moisture proportion, In order to provide a rate of change of weight, As a threshold value for the rate of change of weight, Is the highest temperature of the surface of the clothes, Is the highest safe temperature.
  9. 9. The reinforcement learning-based multi-temperature-zone energy-saving control method according to claim 1, wherein the uniformity index is calculated by the sum of squares of deviation of moisture content of each zone and an average value, and the comprehensive score is calculated by: The calculation is performed such that, Wherein the method comprises the steps of As the weight coefficient of the light-emitting diode, Is the total energy consumption.
  10. 10. The reinforcement learning-based multi-temperature zone energy saving control method according to claim 1, wherein the process of parameter self-optimization comprises: Defining a state space s to include the material type and the initial moisture proportion Uniformity of moisture distribution The motion space a contains a temperature gradient coefficient Coefficient of wind speed Time reference coefficient The calculation formula of the updated action value is as follows: , Wherein the method comprises the steps of For the learning rate, gamma is the discount factor, In order to obtain the final moisture proportion and final moisture distribution uniformity, Is in a state of A set of all actions that may be performed next, S is comprehensive grading; The optimal parameter adjustment quantity is determined by the Q value, and the formula is as follows: ; The final optimization parameter formula is: , Wherein the method comprises the steps of In order to finally optimize the parameters, Lambda is an adjustment step length coefficient for the last optimization parameter; The process for updating the knowledge base comprises the following steps: and storing parameters for successful cases according to similarity weighting, wherein a case weight calculation formula is as follows: , Wherein the method comprises the steps of In order to be a historical case state, In the event of a current state, Is a similarity coefficient; the calculation formula of the error value of the failure case is as follows: , Wherein the method comprises the steps of For the actual composite score to be a function of, Scoring for a prediction synthesis; When (when) When the model parameter calibration is triggered, the calibration formula is as follows: , Wherein the method comprises the steps of As a parameter of the model, it is possible to provide, In order to calibrate the coefficients of the light-emitting diode, Gradient as a loss function; The reference parameters are updated periodically by means of a moving average, and the formula is as follows: , Wherein the method comprises the steps of In order to smooth the coefficient of the coefficient, And the optimal parameter is t time.

Description

Multi-temperature-zone energy-saving control method based on reinforcement learning Technical Field The invention relates to the technical field of automatic control, and particularly discloses a multi-temperature-zone energy-saving control method based on reinforcement learning. Background Under the background of rapid development of intelligent household appliance technology, the drying equipment is used as important household equipment, and the requirements of energy conservation and drying accuracy are increasingly improved. The temperature zone regulation and control of the traditional drying equipment mainly depends on temperature sensing by a temperature sensor and an infrared sensor, and a certain energy-saving effect is realized by dynamically adjusting the drying state. The prior art has obvious limitations that, on one hand, the multi-dimensional state of a controlled object cannot be converted into control input by only relying on a single temperature sensor, so that a control model is disjointed from the state of an actual object, on the other hand, a closed-loop control framework of state sensing-parameter adjustment-effect feedback is not formed by temperature region parameter adjustment depending on a preset fixed rule, and the parameters cannot be corrected in real time according to dynamic change of the controlled object, and furthermore, the algorithm optimization capability based on historical control data is lacking, the control precision and the energy efficiency ratio are difficult to continuously improve, and particularly, the problem of excessive control or insufficient control is easy to occur in a scene of dynamic change of a load. Therefore, it is important to develop an energy-saving regulation and control method of drying equipment, which can accurately sense moisture distribution, realize self-adaptive regulation and control in multiple temperature areas and has self-learning capability. Disclosure of Invention The multi-temperature-zone energy-saving control method based on reinforcement learning is characterized by comprising the following steps of: S1, acquiring initial weight and current weight of clothes through a high-precision weighing sensor, and calculating to obtain integral moisture proportion; Acquiring the moisture content distribution of different areas of the clothes through a microwave moisture sensor array, drawing a two-dimensional thermodynamic diagram to form a moisture distribution map, calculating the average value of the moisture of the areas, and marking a high-humidity area with the moisture content higher than the average value; acquiring the temperature distribution of the surface of the clothes by using a thermal infrared imager, and identifying the type of the clothes material by using near infrared spectrum analysis; integrating the data to construct a controlled object state model comprising the whole moisture proportion, the moisture distribution map, the high humidity area and the material type; S2, dividing the drying cabin into a plurality of independent temperature areas by a multi-temperature-area self-adaptive control strategy, determining basic temperature gradients of the temperature areas through preset rules based on the types of clothes materials, calculating temperature deviations of the temperature areas according to a moisture distribution map and a high-humidity area to adjust initial temperature parameters, calculating moisture change rate according to an average value of the moisture of the area, and self-adaptively controlling power of the temperature areas; S3, optimizing and learning a drying process, monitoring energy consumption of each temperature area in real time, calculating an overall energy efficiency ratio, dynamically adjusting a drying strategy according to the energy efficiency ratio, and judging whether drying is finished or not based on preset main judging conditions, auxiliary judging conditions and safety judging conditions; and S4, post-evaluation and system learning, namely comprehensively detecting the final humidity distribution of the clothes, calculating uniformity indexes, total energy consumption and total time, obtaining comprehensive scores, performing parameter self-optimization based on reinforcement learning, and updating a knowledge base. Preferably, the multi-sensor comprises a high-precision weighing sensor, a microwave moisture sensor array, a thermal infrared imager and a near infrared spectrum analyzer; wherein, the high-precision weighing sensor is used for collecting initial weight And the current weightThe microwave moisture sensor array is used for collecting moisture content distribution of different areasThe thermal infrared imager is used for collecting the temperature distribution of the surface of the clothesThe near infrared spectrum analyzer is used for identifying the material type of the clothes. Preferably, the overall moisture ratio is calculated by the formula: And calculatin