CN108549232B - A kind of room air self-adaptation control method based on approximate model planning - Google Patents
A kind of room air self-adaptation control method based on approximate model planning Download PDFInfo
- Publication number
- CN108549232B CN108549232B CN201810430729.0A CN201810430729A CN108549232B CN 108549232 B CN108549232 B CN 108549232B CN 201810430729 A CN201810430729 A CN 201810430729A CN 108549232 B CN108549232 B CN 108549232B
- Authority
- CN
- China
- Prior art keywords
- state
- model
- parameter
- function
- vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Air Conditioning Control Device (AREA)
Abstract
The invention discloses a kind of room air self-adaptation control methods based on approximate model planning, including initialization current state, model, hyper parameter, environment and explore strategy, it is executed according to policy selection movement is explored, to obtain award and NextState, current state, movement, award and NextState are formed into current sample with more new model, value function and strategy;The sample track of present sample track and reconstruct is added in the pond of track after each plot, then using all tracks in the pond of track come more new model;Analog sample is generated using the model of update to be planned;When algorithm reaches maximum plot number and restrains, so that it may the optimal policy of accomplished room air self adaptive control.The present invention is planned by one approximate environmental model of study using the environmental model of study, to improve the efficiency of study.
Description
Technical field
The present invention relates to a kind of room air self-adaptation control method, more particularly to a kind of based on approximate model planning
Room air self-adaptation control method.
Background technique
With the economic development and improvement of people's living standards, people are also growing day by day for the attention degree of environment.
A place of the indoor environment as people's main activities, it is more close with people's health, therefore, how to effectively realize room
Interior air it is safe, pure and fresh and comfortable, be improve mankind's self a critical issue.
In current most indoor environment, it is only mounted with the equipment such as air-conditioning and air purifier, and be between them
It is isolated existing, it needs individually to come starting device and realizes to the adjusting of air themperature and the purification of air.For some danger
Gas cannot achieve quickly to adjust by air purifier and rapidly will merely such as the formaldehyde and sulfur dioxide in confined space
Its indoor content is reduced within secure threshold, and needing to open a window immediately carries out convection current.Therefore, it is necessary to increase window control equipment.
However, the equipment such as air-conditioning and air purifier require manual control and adjustment, do not have intelligence.Therefore it needs to design corresponding
Control method enable these equipment to start and operate automatically by perception environment, realize to indoor environment in real time from
Dynamic control.
Summary of the invention
For above-mentioned prior art defect, task of the invention lies in provide a kind of Interior Space based on approximate model planning
Gas self-adaptation control method realizes automatic control terminal to meet indoor air environment requirement, while realizing as much as possible indoor
The maximum comfort and satisfaction of personnel.
The technical scheme is that such: a kind of room air self-adaptation control method based on approximate model planning,
The following steps are included:
Step 1), initialization Markovian decision model, are arranged the state space X and motion space U of environment;
Step 2), initiation parameter vector, the parameter vector include: value function parameterPolicing parameter
State transition function parameterReward functions parameterWith eligibility trace parameter
The hyper parameter of step 3), initialization algorithm, the hyper parameter include: discount rate γ, decay factor λ, plot number E,
The learning rate α of maximum time step S, value function that the exploration standard deviation ε of Gaussian function, each plot are included1, strategy study
Rate α2, the learning rate α of model and the number K of planning;
Step 4), initialization current episode s=1;
Step 5), initialization current state xt=x, current time walk t=1;
Step 6), selection movement: according to the movement u that should be executed under exploration policy selection current statet=u;
Step 7) generates sample: in current state xtUnder, execution acts u, obtains next state xt+1It awards immediately
rt+1, the sample of generation is (xt,ut,xt+1,rt+1);
Step 8), using sample (xt,ut,xt+1,rt+1) more new model state transition function parameter vector θ and award letter
Number parameter vector υ;
Step 9) calculates time difference error (Temporal Difference, TD) error;
Step 10) updates eligibility trace: updating eligibility trace parameter vector;
Step 11) updates value function: updating the corresponding parameter vector of value function;
Step 12), more new strategy: the corresponding parameter vector of more new strategy;
Step 13) updates current state: xt=xt+1;
Step 14) updates current time step t=t+1, judges whether to reach maximum time step: if reached, being transferred to step
15);Otherwise, step 5) is transferred to continue to execute;
Step 15) is planned using approximate model;
Step 16) updates current episode s=s+1, judges whether to reach maximum plot number: if reached, being transferred to step
17);Otherwise, step 5) is transferred to continue to execute;
Step 17), the optimal policy for obtaining room air self adaptive control to optimal policy according to study.
Value function approximate representation described in step (2) is as a preferred technical solution,Wherein,For Gaussian function, for state x to be mapped as feature vector,Centered on point, σ1For the standard of state dimension
Difference, ξ are parameter vector, and dimension is consistent with feature vector;It is described strategy approximate representation beIt is wherein special
Levy vectorIdentical as the expression of value function, ζ is policing parameter vector;Model includes state transition function and reward functions,
Migration approximation to function is expressed asReward functions approximate representation is rt+1=φT(xt,ut)υt, whereinFor state action feature,For the central point of movement, σ2For the standard deviation for acting dimension, θ is state
The parameter vector of function is migrated, υ is the parameter vector of reward functions.
The exploration strategy in step (6) is generated using Gaussian function as a preferred technical solution, in free position place
The movement taken according toIt obtains, wherein h (x)=u*Expression obtains most at state x according to optimal policy
Excellent movement, ε are to explore the factor.
The more new model in step (8) is using the prediction error of single step as gradient signal as a preferred technical solution:
By the status predication error of single stepAs gradient, obtaining migration function parameter vector more new formula isError is predicted into the award of single stepAs gradient, parameter vector is obtained
More new formula is
The calculation formula of calculating TD error in step (9) as a preferred technical solution, are as follows: ω=r+ γ V (xt+1)-V
(xt)。
Eligibility trace more new formula in step (10) as a preferred technical solution, are as follows:
Step (11) median function more new formula as a preferred technical solution, are as follows: ξt+1=ξt+α1ωe(xt)。
Policy update formula in step (12) as a preferred technical solution, are as follows: ζt+1=ζt+α2ω(u-u*)Tφ(xt)。
The model planning in step (15) is the iteration land productivity under certain planning number as a preferred technical solution,
With model xt+1=φT(xt,ut)θtAnd rt+1=φT(xt,ut)υtNext state and award are generated, and utilizes the state and prize
Reward carrys out updated value function parameter vector ξt+1=ξt+α1ωe(xt) and policing parameter vector ζ tt+1=ζt+α2ω(u-u*)Tφ
(xt)。
The advantages of the present invention over the prior art are that: the nitrification enhancement based on approximate model planning passes through
An approximate environmental model is practised, and carries out sector planning using the environmental model of study, to improve the efficiency of study.It is logical
Continuous acquisition terminal equipment (air purifier and air-conditioning) is crossed from the perception data on belt sensor to learn optimal policy, is utilized
The optimal policy learnt controls equipment to start accordingly, starts corresponding terminal device (window, air purifier and sky
Adjust) indoor environment automatically controlled in real time.
Detailed description of the invention
Fig. 1 is that the present invention is based on the room air adaptive control system structural schematic diagrams that approximate model is planned;
Fig. 2 is server architecture schematic diagram in room air adaptive control system of the present invention;
Fig. 3 is the whole control flow signal for the room air adaptive control system planned the present invention is based on approximate model
Figure;
Fig. 4 is that the present invention is based on the room air self-adaptation control method flow diagrams that approximate model is planned.
Specific embodiment
Below with reference to embodiment, the invention will be further described, but not as a limitation of the invention.
A kind of room air self-adaptation control method based on approximate model planning that the present embodiment is related to is applied to such as Fig. 1
Shown in room air self-adapted adjustment system, the main modular in the system has: 1, air purifier 2, air-conditioning 3, window control
Control equipment 4, clarifier control equipment 5, air-conditioning control equipment 6, display and management equipment 7, server and 8 cell phone applications.Each module
Between can be used wireless network connection, the organizational form of wireless network uses WI-FI network, but also can choose GPRS, 3G, 4G
And Zigbee is as cordless communication network.Incorporated by reference to shown in Fig. 2, the chief component of server have central controller 8a,
Storage unit 8b, sensor unit 8c include temperature sensor, humidity sensor, formaldehyde sensor, sulfur dioxide sensing
The various sensings such as device, PM2.5 sensor (done expansion interface, can be convenient the new sensor type of increase in this part of sensor)
The interface circuit 8d of device and all kinds of control equipment.In conjunction with shown in Fig. 3, cycle sensor sends data to server,
Server after receiving the data, through current value compared with secure threshold, come determine based on approximate model planning reinforcing
The reward value of learning algorithm, and the nitrification enhancement planned based on approximate model is inputted using the data as sample data, it learns
Commonly use the optimal policy in control.In addition, issuing control life to control equipment when the data value of acquisition is more than secure threshold
It enables, while prompt indoor occupant is sounded an alarm by display and management equipment, house-owner and tenant are sent information to by internet
Cell phone application on, carry out the prompt of relevant information.
Room air self-adaptation control method based on approximate model planning mainly includes two aspects:
Firstly, it is necessary to be judged data and formatted:
1) temperature sensor: setting normal temperature value should be 18 DEG C~28 DEG C, when temperature is located at this section, temperature
Value is normal value;
2) humidity sensor: normal humidity value is set as 40%~60%, when humidity is located at this section, humidity value
For normal value;
3) formaldehyde sensor: setting normal formolite number as 0~0.08mg/m3, when content of formaldehyde is located at this section,
Formolite number is normal value;
4) SO 2 sensor: normal titanium dioxide sulfur number is set as 0~0.50mg/m3, when sulfur dioxide concentration position
When this section, titanium dioxide sulfur number is positive constant value;
5) PM2.5 sensor: normal PM2.5 value is set as 0~75ug/m3, when PM2.5 concentration is located at this section
When, PM2.5 belongs to normal value.
When the data that above-mentioned value sensor is sent are normal value, control equipment is without carrying out any operation;When it
In certain data when occurring abnormal, it is necessary to send and order to corresponding control equipment, starting relevant device progress corresponding operating.
Such as when the concentration of sulfur dioxide is more than 0.5mg/m3, indoor concentration is more than safety value, can send commands to window control equipment,
To which startup trigger opens window.
Error is calculated in order to exclude different data dimension bring, Regularization is carried out to each data, using formulaTo be handled, wherein xmaxIndicate maximum value, xminIndicate minimum value,Indicate current value, then regularization
Value afterwards byX is converted to, range is between [0,1].
Secondly, optimal policy is solved using the nitrification enhancement planned based on approximate model, in order to realize room air
Automatic real-time control, need for the control problem of indoor control to be modeled as a MDP problem first, then recall corresponding
Algorithm solved.Room air adaptive control system is modeled rear corresponding MDP and can indicate are as follows:
(1) state space: the dimension of state is 5, and the component of state mainly includes temperature sensor, humidity sensor, first
The reading of aldehyde sensor, SO 2 sensor and PM2.5 sensor.In state space free position x=temperature, it is wet
Degree, formaldehyde, sulfur dioxide, PM2.5 };
(2) motion space: the dimension of movement is 3, can be expressed as u={ movement of air-conditioning control equipment, air purifier
Control the movement of equipment, the movement of window control equipment }.
The movement of air-conditioning control equipment specifically includes that the small wind of 1 refrigeration, 2 refrigeration strong wind, the small wind of 3 heatings, 4 heating strong wind, 5
Humidification, 6 dehumidifying.
The movement of air purifier control equipment specifically includes that 1 purification, 2 are closed.
The movement of window control equipment specifically includes that 1 maximum angle opens (90 °), 2 wide-angles opening (is less than greater than 60 °
90 °), 3 moderate angles open (be greater than 30 ° less than 60 °), 4 low-angles are opened and (are greater than 0 ° less than 30 °), 3 close.
(3) reward functions: reward functions are to be manually set, and the value having can be according to risk, the subjective experience of people
The operational effect of skewed popularity and algorithm is finely adjusted, such as to this kind of hazardous gas of sulfur dioxide, it will usually be arranged one it is larger
Negative reward so that controller learns optimal policy in this state as early as possible, secondly, if indoor occupant is to temperature high nothing
Method is endured, and can give a larger negative reward when temperature value is higher than range of normal value.
In this example implementation process, the design of reward functions is as follows:
1) temperature value: when measured temperature is at normal value section, otherwise it is -5 that reward value, which is+1,;
2) humidity value: when humidity measurements are at normal value section, otherwise it is -1 that reward value, which is+1,;
3) formolite number: when formaldehyde measurement value is at normal value section, otherwise it is -10 that reward value, which is+1,;
4) titanium dioxide sulfur number: when sulfur dioxide measured value is at normal value section, otherwise it is -15 that reward value, which is+1,;
5) PM2.5 value: when PM2.5 measured value is at normal value section, otherwise it is -8 that reward value, which is+1,;
(4) migrate function: after migration function representation executes the movement of selection under current state, system or environment moved
The next state moved on to.Since the state in the example is realized by reading sensing data, move to next
A state can also be obtained by reading the data of sensor.
It is illustrated in figure 4 the process of the nitrification enhancement based on approximate model planning in control centre, the tool executed
Body process includes below step:
Step 1), initialization Markovian decision model, i.e., be modeled as one for room air control problem according to above-mentioned
MDP problem, init state space, motion space, reward functions and migration function;
Step 2), initiation parameter vector, parameter vector mainly includes: value function parameterPolicing parameter
State transition function parameterReward functions parameterWith eligibility trace parameter
The hyper parameter of step 3), initialization algorithm, hyper parameter specifically include that setting discount rate γ=0.95, decay factor λ
=0.85, plot number E=500, exploration standard deviation ε=0.6 of Gaussian function, the maximum time that each plot is included walk S=
400, the learning rate α of value function1=0.7, tactful learning rate α2=0.6, learning rate α=0.5 of model, the number K=of planning
100;
Step 4), initialization current episode s=1;
Step 5), initialization current state xtThe initial reading of=each sensor, current time walk t=1;
Step 6), selection movement: the movement that should be executed under current state is selected according to strategy is explored
Step 7) generates sample: in current state xtUnder, execution movement u (does not execute any movement or starting control equipment
Responded), the reading for reading sensor obtains next state xt+1R is awarded immediatelyt+1, the sample of generation is (xt,ut,
xt+1,rt+1);
Step 8), learning model: sample (x is utilizedt,ut,xt+1,rt+1) come state transition function and the award of more new model
The parameter vector of functionWith
Step 9) calculates TD error: calculating TD error ω=r+ γ V (x firstt+1)-V(xt);
Step 10) updates eligibility trace: updating eligibility trace parameter vector
Step 11) updates value function: updating the corresponding parameter vector ξ of value functiont+1=ξt+α1ωe(xt);
Step 12), more new strategy: the corresponding parameter vector ζ t of more new strategyt+1=ζt+α2ω(u-u*)Tφ(xt);
Step 13) updates current state: saving the reading x of current sensort=xt+1;
Step 14) updates current time step t=t+1, judges whether to reach maximum time step: if reached, being transferred to step
15);Otherwise, step 5) is transferred to continue to execute;
Step 15) plans that initialization current state is current ambient conditions, and initializes eligibility trace using approximate modelCirculation executes K times, according toSelection movement;Calculate next state x of predictiont+1=φT(xt,ut)
θt;Calculate the award r of predictiont+1=φT(xt,ut)υt;Establish analog sample (xt,ut,xt+1,rt+1);Calculate TD error ω=r+
γV(xt+1)-V(xt);Update eligibility trace parameterUpdated value function parameter ξt+1=ξt+α1ωe1
(xt);Update policing parameter ζt+1=ζt+α2ω(u-u*)Tφ(xt);
Step 16) updates current episode s=s+1, judges whether to reach maximum plot number: if reached, being transferred to step
17);Otherwise, step 5) is transferred to continue to execute;
Step 17), the optimal policy for obtaining indoor environment self adaptive control to optimal policy according to study.
Claims (3)
1. a kind of room air self-adaptation control method based on approximate model planning, which comprises the following steps:
Step 1), initialization Markovian decision model, are arranged the state space X and motion space U of environment;
Step 2), initiation parameter vector, the parameter vector include: value function parameterPolicing parameterState
Migrate function parameterReward functions parameterWith eligibility trace parameter
The hyper parameter of step 3), initialization algorithm, the hyper parameter include: setting discount rate γ, decay factor λ, plot number E,
The exploration standard deviation ε of Gaussian function, the maximum time that each plot is included walk S, the learning rate α of value function1, tactful study
Rate α2, the learning rate α of model, the number K of planning;
Step 4), initialization current episode s=1;
Step 5), initialization current state xt=x, current time walk t=1;
Step 6), selection movement: according to the movement u that should be executed under exploration policy selection current statet=u;The exploration strategy is adopted
Generated with Gaussian function, the movement taken at free position according toIt obtains, wherein h (x)=u*It indicates
The optimal movement obtained at state x according to optimal policy;
Step 7) generates sample: in current state xtUnder, execution acts u, obtains next state xt+1R is awarded immediatelyt+1, raw
At sample be (xt,ut,xt+1,rt+1);
Step 8), using sample (xt,ut,xt+1,rt+1) more new model state transition function parameter vector θ and reward functions ginseng
Number vector υ, the more new model is using the prediction error of single step as gradient signal: by the status predication error of single stepAs gradient, obtaining migration function parameter vector more new formula isIt will be single
Error is predicted in the award of stepAs gradient, the more new formula for obtaining reward functions parameter vector isφ(xt,ut) it is state action feature;
Step 9) calculates TD error;
Step 10) updates eligibility trace: updating eligibility trace parameter vector, eligibility trace more new formula is For the corresponding feature vector of state x;
Step 11) updates value function: updating the corresponding parameter vector of value function, value function more new formula is ξt+1=ξt+α1ωe
(xt);
Step 12), more new strategy: the corresponding parameter vector of more new strategy, policy update formula are ζt+1=ξt+α2ω(u-u*)T
φ(xt), φ (xt) it is state feature;
Step 13) updates current state: xt=xt+1;
Step 14) updates current time step t=t+1, judges whether to reach maximum time step: if reached, being transferred to step 15);
Otherwise, step 5) is transferred to continue to execute;
Step 15) plans that the approximate model planning is iteratively utilized under certain planning number using approximate model
Model xt+1=φT(xt,ut)θtAnd rt+1=φT(xt,ut)υtNext state and award are generated, and utilizes the state and award
Carry out updated value function parameter vector ξt+1=ξt+α1ωe(xt) and policing parameter vector ζt+1=ζt+α2ω(u-u*)Tφ(xt);
Step 16) updates current episode s=s+1, judges whether to reach maximum plot number: if reached, being transferred to step 17);It is no
Then, step 5) is transferred to continue to execute;
Step 17), the optimal policy for obtaining room air self adaptive control to optimal policy according to study.
2. the room air self-adaptation control method according to claim 1 based on approximate model planning, which is characterized in that
Value function approximate representation described in step (2) isWherein, Gaussian functionFor by state x
Feature vector is mapped as,Centered on point, σ is the standard deviation of Gaussian function, and dimension and the feature vector of ξ be consistent;It is described
Tactful approximate representation isWherein, feature vectorIt is identical as the expression of value function;Model is moved comprising state
Function and reward functions are moved, migration approximation to function is expressed as xt+1=φT(xt,ut)θt;Reward functions can be by approximate representation
rt+1=φT(xt,ut)υt, whereinFor state action feature,For the central point of movement, σ1For state
The standard deviation of dimension, σ2For the standard deviation for acting dimension, θ is the parameter vector of state transition function, and υ is the parameter of reward functions
Vector.
3. the room air self-adaptation control method according to claim 1 based on approximate model planning, which is characterized in that
The calculation formula of calculating TD error in step (9) are as follows: ω=r+ γ V (xt+1)-V(xt)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810430729.0A CN108549232B (en) | 2018-05-08 | 2018-05-08 | A kind of room air self-adaptation control method based on approximate model planning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810430729.0A CN108549232B (en) | 2018-05-08 | 2018-05-08 | A kind of room air self-adaptation control method based on approximate model planning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108549232A CN108549232A (en) | 2018-09-18 |
CN108549232B true CN108549232B (en) | 2019-11-08 |
Family
ID=63513603
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810430729.0A Active CN108549232B (en) | 2018-05-08 | 2018-05-08 | A kind of room air self-adaptation control method based on approximate model planning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108549232B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110531620B (en) * | 2019-09-02 | 2020-09-18 | 常熟理工学院 | Adaptive control method of mountain climbing system of trolley based on Gaussian process approximate model |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103942575A (en) * | 2014-04-02 | 2014-07-23 | 公安部第三研究所 | System and method for analyzing intelligent behaviors based on scenes and Markov logic network |
CN105740644B (en) * | 2016-03-24 | 2018-04-13 | 苏州大学 | Cleaning robot optimal target path planning method based on model learning |
CN107065582B (en) * | 2017-03-31 | 2023-09-29 | 苏州科技大学 | Indoor air intelligent adjusting system and method based on environment parameters |
CN107272785B (en) * | 2017-07-19 | 2019-07-30 | 北京上格云技术有限公司 | A kind of electromechanical equipment and its control method, computer-readable medium |
-
2018
- 2018-05-08 CN CN201810430729.0A patent/CN108549232B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN108549232A (en) | 2018-09-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108900346A (en) | Wireless network traffic prediction technique based on LSTM network | |
CN103912966B (en) | A kind of earth source heat pump refrigeration system optimal control method | |
CN108105844B (en) | Intelligent heat exchange station control method based on indoor and outdoor temperature combined compensation | |
CN100501608C (en) | Posterior control and multi-sensor infusion method and central conditioner energy-saving control device | |
CN112434787B (en) | Terminal space energy consumption prediction method, medium and equipment based on total energy consumption of building | |
CN108399470B (en) | Indoor PM2.5 prediction method based on multi-example genetic neural network | |
CN107065582A (en) | The Intelligent Indoor Air regulating system and adjusting method of a kind of parameter based on environment | |
CN109165798A (en) | A kind of Free Calcium Oxide Contents in Cement Clinker on-line prediction method and system | |
CN112413831A (en) | Energy-saving control system and method for central air conditioner | |
CN108827386A (en) | Building model indoor natural ventilation testing device and testing method | |
CN112947648B (en) | Agricultural greenhouse environment prediction method and system | |
CN115907366B (en) | Agricultural product growth environment optimal regulation and control method and equipment based on flamingo algorithm | |
CN111103220A (en) | Method and system for predicting and regulating concentration of atmospheric pollutants | |
Yue et al. | The prediction of greenhouse temperature and humidity based on LM-RBF network | |
CN112577159A (en) | Air conditioner energy-saving intelligent control method and device based on human body thermal comfort | |
CN108549232B (en) | A kind of room air self-adaptation control method based on approximate model planning | |
CN110097929A (en) | A kind of blast furnace molten iron silicon content on-line prediction method | |
CN105134632A (en) | Intelligent electric fan and method for blowing shading area | |
CN116729060A (en) | DDPG-based pure electric vehicle passenger cabin air conditioner refrigeration control method | |
Zhuang et al. | Probabilistic occupancy forecasting for risk-aware optimal ventilation through autoencoder Bayesian deep neural networks | |
Tomažič et al. | Indoor-environment simulator for control design purposes | |
KR102166146B1 (en) | Apparatus and Method for Controlling Multiple windows in a greenhouse | |
CN114110824B (en) | Intelligent control method and device for constant humidity machine | |
Cicirelli et al. | Balancing energy consumption and thermal comfort with deep reinforcement learning | |
CN106403162A (en) | Local heat comfort control method, local heat comfort controller and control system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |