Building environment carbon reduction system based on artificial intelligence behavior deep learning

Abstract

The invention relates to the technical field of low-carbon buildings, in particular to a building environment carbon reduction system based on artificial intelligence behavior deep learning. Comprising an environmental carbon judgment module, an equipment execution module and a carbon behavior deep learning module, the input of the environmental carbon judgment module is first data, the environmental carbon judgment module converts the input first data into character string parameters, and the input of the equipment execution module is character string parameters; the equipment execution module processes an input character string parameter and then outputs a control instruction and a behavior message queue, the input of the carbon behavior deep learning module is the behavior message queue, and the carbon behavior deep learning module processes the behavior message queue to obtain a first coding file and a second coding file; updating the environment carbon judgment module according to the first coding file, and updating the equipment execution module according to the second coding file; the computing resource consumption is reduced, the artificial intelligence coverage range is improved, and the adaptability of artificial intelligence to actual application in different environments is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of low-carbon buildings, and particularly to a building environment carbon reduction system based on artificial intelligence behavior deep learning. Background Art

[0002] Low-carbon building operation and maintenance aims to reduce carbon emissions throughout the building life cycle, improve energy utilization efficiency, and further enhance the green and sustainable development ability during the building use and operation and maintenance processes. With the increasing global emphasis on sustainable development and environmental protection, the construction industry (especially the use, operation, and maintenance processes of buildings as the main activity places for humans) as an important field of energy consumption and carbon emissions is facing the problem of transformation towards low-carbon and intelligent development.

[0003] In the actual building operation and maintenance process, due to the existence of human activities, and the building as an internal environment provider to meet the daily activity needs of humans, an internal environment different from the outdoor environment needs to be provided, so the energy consumption cannot reach zero. However, through relevant monitoring methods, the energy consumption can be reduced by optimizing the building energy distribution and usage methods, thereby achieving the purpose of low-carbon building operation. However, the existing monitoring methods have high computational resource consumption and low artificial intelligence coverage. Also, due to the extremely large individual differences in the building internal environment (in the same building, for the same office use, different companies will create completely different building internal environments due to individual differences and preferences), the existing artificial intelligence has poor adaptability to this spatial physical differentiation, resulting in poor actual application effects.

[0004] Therefore, there is an urgent need to provide a building environment carbon reduction system based on artificial intelligence behavior deep learning. Summary of the Invention

[0005] The present invention solves the technical problems existing in the prior art, and provides a building environment carbon reduction system based on artificial intelligence behavior deep learning.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A building environment carbon reduction system based on artificial intelligence behavior deep learning, including an environmental carbon judgment module, a device execution module, and a carbon behavior deep learning module. The input of the environmental carbon judgment module is the first data. The environmental carbon judgment module converts the input first data into string parameters and outputs them. The input of the device execution module is the string parameters output by the environmental carbon judgment module. The device execution module processes the input string parameters and outputs a control instruction and a behavior message queue. The input of the carbon behavior deep learning module is the behavior message queue. The carbon behavior deep learning module processes the behavior message queue to obtain a first coding file and a second coding file, and updates the environmental carbon judgment module according to the first coding file and updates the device execution module according to the second coding file.

[0008] Further, the environmental carbon judgment module includes a data collection and conversion module, a judgment parameter input and conversion module, a first judgment module, and a parameter output module;

[0009] The data collection and conversion module is used to obtain the first data, where the first data includes the environmental temperature TemE and the power system power. The data collection and conversion module removes the green energy part from the power system power to obtain the carbon-consuming power. The judgment parameter input and conversion module is used to construct a time data list, obtain the set temperature TemS, the set time, and the sensitive parameter K. The first judgment module is used to judge the input carbon-consuming power, environmental temperature, set temperature, set time, and sensitive parameter to obtain a return value. The parameter output module is used to form string parameters according to the return value processed by the first judgment module.

[0010] Even further, in the first judgment module, the specific method for obtaining the return value is as follows:

[0011] The input data is corrected using the TemK(TemE, K) algorithm. First, the data is screened, and a constant k is set. According to the set constant k, adjacent data with an absolute value of the environmental temperature change less than |k| is merged to obtain a merged data set. The merged data set includes multiple merged data pairs (TemEn, kn), where TemEn represents the nth environmental temperature value in the merged data set, and kn represents the time value corresponding to TemEn. A judgment function model is set, with the function variables being TemEn and kn. In the judgment function model, the sigmoid function is used as the activation function, and the environmental temperature in the merged data set is judged through this function. The judged environmental result temperature is denoted as TemEf. A comparison function model is set, and the set temperature TemS in the first data and the environmental result temperature TemEf processed by the judgment function model are input into the comparison function model to calculate the corresponding judgment temperature difference Mnum value. The judgment temperature difference Mnum value is specifically calculated according to the following formula:

[0012] TemEf - TemS = Mnum;

[0013] In the above formula, Mnum represents the judged temperature difference, and TemS represents the set temperature in the first data;

[0014] Input Mnum and the corresponding M value into the judgment part of the comparison function, where the M value is set to 0 or 1 or 2. The specific method to obtain the return value through the judgment part of the comparison function model is as follows:

[0015] (1) When the M value is 0:

[0016] When Mnum ≥ 2, the corresponding return value is to adjust the refrigeration degree to the high - cold value high$cool;

[0017] When 0 < Mnum < 2, the corresponding return value is to adjust the refrigeration degree to the mid - cold value mid$cool;

[0018] When - 1 < Mnum ≤ 0, the corresponding return value is to adjust the refrigeration degree to the low - cold value low$cool;

[0019] When Mnum ≤ - 1, the corresponding return value is stop;

[0020] (2) When the M value is 1:

[0021] When - Mnum ≥ 2, the corresponding return value is to adjust the heat source output to the high - heat value high$heat;

[0022] When 0 < - Mnum < 2, the corresponding return value is to adjust the heat source output to the medium - heat value mid$heat;

[0023] When - 1 < - Mnum ≤ 0, the corresponding return value is to adjust the heat source output to the low - heat value low$heat;

[0024] When - Mnum ≤ - 1, the corresponding return value is stop;

[0025] (3) When the M value is 2, the corresponding return value is air;

[0026] The string parameter formed by the parameter output module is represented by the following formula:

[0027] return′ + “OCT” + pcs′ + “TEM” + teme;

[0028] In the above formula, return′ represents the integer value corresponding to the return value, "OCT" and "TEM" both represent identification characters, pcs′ represents the octal encoding value corresponding to pcs, pcs represents the carbon consumption power, and teme represents the octal encoding value corresponding to TemEf.

[0029] Furthermore, the device execution module includes an information encoding unit, a judgment condition parameter module, a second judgment module, a control module, a physical execution unit, and a behavior message queue module;

[0030] The information encoding unit inputs a string parameter, processes the string parameter to obtain a first parameter, and the first parameter is a first execution instruction encoding with a judgment identification header information J; stores the first parameter into the message queue list, and the information encoding unit outputs the message queue list; or, the information encoding unit inputs an Action encoding with a REN identification header information, processes the Action encoding with a REN identification header information to obtain a second parameter, and the second parameter includes a second execution instruction encoding with a judgment identification header information h;

[0031] The judgment condition parameter module is used to set comprehensive parameters of human body parameters, radar parameters, and external environment; the control module is used to set an activation period, PCS parameters, AI parameters, and a carbon consumption coefficient; the second judgment module inputs human body parameters, radar parameters, comprehensive parameters of the external environment, AI parameters, PCS parameters, and carbon consumption parameters;

[0032] The control module activates the second judgment module according to the activation period, and the second judgment module processes the information in the message queue list and outputs a first execution command instruction; or, the second judgment module is directly activated according to the received physical behavior instruction, and the second judgment module is immediately activated after receiving the second parameter and outputs a second execution command instruction and a behavior data json file;

[0033] The physical execution unit is used to parse the first and second execution command instructions into an instruction set queue, and mobilize devices according to different instruction sets for queue execution; the behavior message queue module includes multiple behavior data in json file format, and this behavior message queue is a set list of behavior data in json file format for a time period.

[0034] Furthermore, the control module activates the second judgment module according to the control period. When activating, it judges the ai_Model parameter in the AI parameters and processes the first parameter in the message queue list. The specific method is as follows:

[0035] First step, obtain the instructions in the message queue list and input them into the judgment function F. If the return value is null, do not proceed to the next step; otherwise, proceed to the second step.

[0036] Second step, parse teme, pcs, and Act_string1 from the first returned parameter, and judge the carbon consumption change Pcs′ of the entire air conditioning system after Act_string1 is executed according to the carbon consumption parameter. The specific judgment method is as follows:

[0037] If pcs + Pcs′ > PCS, then the intermediate parameter pcsbool = false;

[0038] If pcs + Pcs′ ≤ PCS, then the intermediate parameter pcsbool = true;

[0039] Third step, when pcsbool = true and the ai_Model parameter is 2, proceed to the fourth step; otherwise, do not proceed to the next step and stop.

[0040] Fourth step, obtain the external temperature value. When the temperature value after translating teme is greater than the external temperature value, do not execute all instructions containing heating; when the temperature value after translating teme is less than the external temperature value, do not execute all instructions containing cooling. According to different situations, edit the corresponding instructions in Act_string1 to form an execution command instruction and output it to the physical execution unit.

[0041] Furthermore, in the second judgment module, directly activate according to the behavior, activate when receiving the second parameter, parse the second parameter to obtain the execution parameter Act_string2; then judge the ai_Model parameter:

[0042] When the ai_Model parameter is 0, send the second execution command instruction to the physical execution unit;

[0043] When the ai_Model parameter is 1, send the second execution command instruction to the physical execution unit, and at the same time form the behavior data from the data related to the second execution command instruction, save it as a json file, and send it to the behavior message queue module;

[0044] When the ai_Model parameter is 2, perform the following steps:

[0045] First step, call the last first parameter in the message queue list and parse the teme in it to obtain the external environmental temperature, and then judge. When the temperature value after translating teme is greater than the external environmental temperature value, do not execute all instructions containing heating; when the temperature value after translating teme is less than the external environmental temperature value, do not execute all instructions containing cooling; in other cases, proceed to the second step.

[0046] In the second step, obtain the instructions in the message queue list and input them into the judgment function F. If the return value is null, do not proceed to the next step; otherwise, proceed to the third step.

[0047] In the third step, parse teme, pcs, and Act_string1 from the first returned parameter. According to the carbon consumption parameter C, judge the carbon consumption change Pcs′ of the entire air conditioning system after Act_string1 is executed. The specific judgment method is as follows:

[0048] If pcs + Pcs′ > PCS, stop.

[0049] If pcs + Pcs′ ≤ PCS, trigger the execution command and edit the relevant execution instructions in Act_string1.

[0050] In the fourth step, transfer Act_string1 to the physical execution unit. At the same time, form behavior data from the data related to the execution instructions, save it as a json file, and send it to the behavior message queue module.

[0051] Furthermore, the carbon behavior deep learning module includes a deep learning setting module, a behavior learning module, a refresh module, and a control information setting module.

[0052] The deep learning setting module is provided with an ai_Model parameter. Only when the ai_Model parameter is 0, the behavior learning module is not activated; otherwise, the behavior learning module is activated. The behavior learning module includes a first AI model, a second AI model, and an AI behavior tree. Both the first AI model and the second AI model learn and evolve based on the behavior data and give corresponding parameter adjustment results. The AI behavior tree is responsible for selecting one of the models and adjusting the probability ratio between the two selections, and outputting the PCS value, TemS value, Atime parameter, and K parameter. When the ai_Model parameter is 1, the output value is sent to the refresh module; when the ai_Model parameter is 2, the output value is sent to the control information setting module. When the refresh module receives the output value, it clears all the instruction information in the behavior message queue. When the control information setting module receives the information, it converts the TemS value and the K parameter into strings and encodes them to form a first encoded file, and transports the first encoded file to the environmental carbon judgment module for update; it converts the Atime parameter and the PCS value into strings and encodes them to form a second encoded file, and transports the second encoded file to the device execution module for update; at the same time, the refresh module is activated to clear all the instruction information in the behavior message queue.

[0053] Furthermore, the first AI model is a deep learning model, including a linear layer, a non-linear layer, and an output layer. The linear layer is used to input behavior data. The linear layer uses the transformer method to obtain the Atime parameter, PCS parameter, TemS parameter, and K parameter. The output layer is used to output the Atime parameter, PCS parameter, TemS parameter, and K parameter;

[0054] The second AI model is a reinforcement learning model, which uses the Qlearning method to obtain the output Atime parameter, PCS parameter, and K parameter.

[0055] Furthermore, the AI behavior tree uses the Qlearning method to select the output results of the first AI model and the second AI model according to the weight parameter E, and corrects the weight parameter E according to the adjustment times parameter of the previous cycle, and outputs the recommended PCS value, TemS value, Atim parameter, and K parameter.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] The present invention sets up three parts of modules. Each of the three parts of modules has a complete basic artificial intelligence module and an interaction module, and can be applied separately. Therefore, they can be separately arranged on different terminal devices, so that the algorithm has the characteristics of being distributed in physical space. After information connection, the three independent algorithms can be fused to jointly form a deep learning algorithm for carbon reduction artificial intelligence behavior, forming a complete algorithm structure. At this time, the overall algorithm will be executed sequentially between different devices according to a specific algorithm process, and a complete deep learning algorithm for behavior is realized by combining the functional modules between different devices. This distributed layout enables the algorithm to penetrate into various control terminals, can fully expand the coverage of the algorithm, and at the same time greatly enhances the adaptability and system security of the algorithm. The algorithm no longer depends on a specific device or a specific system structure to be realized. The algorithm consists of many independent parts, which means that the absence of any part will not affect the basic operation of the terminal device and the artificial intelligence function. It realizes reducing the consumption of computing resources, increasing the coverage of artificial intelligence, and improving the adaptability and effect of artificial intelligence in actual applications in different environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is the overall flowchart of the method of the present invention.

[0059] Figure 2 is the structural schematic diagram of the environmental carbon judgment module of the present invention.

[0060] Figure 3 is the structural schematic diagram of the device execution module of the present invention.

[0061] Figure 4 is the structural schematic diagram of the carbon behavior deep learning module of the present invention. Detailed implementation manners

[0062] The technical solution of the present invention will be clearly described below in conjunction with the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0063] As Figure 1 shown, the present invention provides a building environment carbon reduction system based on artificial intelligence behavior deep learning, including an environmental carbon judgment module, a device execution module, and a carbon behavior deep learning module; the input of the environmental carbon judgment module is the first data, and the environmental carbon judgment module converts the input first data into string parameters for output; the device execution module inputs the string parameters output by the environmental carbon judgment model, and the device execution module processes the input string parameters and outputs a control instruction and a behavior message queue; the carbon behavior deep learning module inputs the behavior message queue, and the carbon behavior deep learning module outputs a first coding file and a second coding file, so as to update the environmental carbon judgment module and the device execution module.

[0064] The environmental carbon judgment module includes a data collection and conversion module, a judgment parameter input and conversion module, a first judgment module, a parameter processing module, and a parameter output module.

[0065] In the data collection and conversion module, the first data is obtained through corresponding sensing facilities or a sensing network. The first data can be obtained through sensors or transmitted through a collector via relevant protocols (such as existing air temperature collection devices or electricity meter devices). The first data includes environmental temperature and power system power. The environmental temperature is represented as TemE, and the power system power is represented as ps. Then, the green energy part is removed from the power system power to obtain the carbon consumption power, which is specifically removed by the following formula:

[0066] pcs = ps × (1 - G);

[0067] In the above formula, pcs represents the carbon consumption power, and G represents the proportion of green electricity in the power grid where the detection environment is located.

[0068] In the judgment parameter input and conversion module, a time data list is set. The time data list includes temperature data within 10 days in the region where it is located. Specifically, there are a total of 37 temperature data. The time data list can also include relevant data such as the cooling capacity or air conditioning power of the air conditioning system within 10 days; the present invention preferably sets a time data list of temperature data. The temperature data in the time data list is represented as the set temperature TemS, and each set temperature corresponds to a set time TimeNow.

[0069] In the first judgment module, based on the carbon consumption power and the environmental temperature TemE obtained from the data acquisition and conversion module, as well as the set temperature TemS, set time, and sensitive parameters obtained from the judgment parameter input and conversion module, a judgment is made to obtain the return value. The specific method is as follows:

[0070] The input data is corrected. The correction specifically uses the TemK(TemE, K) algorithm for correction. First, the data is screened. A constant k is set. According to the set constant k, adjacent data with an absolute value of temperature change less than |k| are merged. The specific method for the TemK algorithm to correct the input data is as follows: Set the sensitive parameter K, and set the sensitive parameter as the trigger frequency. K is preferably 300s. The environmental temperatures in the input data within each K cycle are merged. Specifically: Set a constant k, compare the absolute value of the change in adjacent environmental temperatures within each K cycle with the absolute value of the set constant |k|, and merge the environmental temperatures less than |k|. The merging method is as follows: Compare the second environmental temperature with the first environmental temperature. When the absolute value of the difference between the second environmental temperature and the first environmental temperature is greater than the absolute value of the set constant |k|, record the first environmental temperature and the second environmental temperature. When the absolute value of the difference between the second environmental temperature and the first environmental temperature is less than or equal to the absolute value of the set constant |k|, record that the temperature value of the second environmental temperature is the same as that of the first environmental temperature. Then compare the third environmental temperature with the second environmental temperature, and so on, to record all environmental temperatures, which can avoid recording a large number of tiny change data with little practical significance. After this step of processing, a merged data set is obtained. The merged data set includes multiple merged data pairs (TemEn, kn), where TemEn represents the nth environmental temperature value in the data set, and kn represents the time value corresponding to TemEn.

[0071] Set a judgment function model. The function variables are TemEn and kn. In the judgment function model, the sigmoid function is used as the activation function. All TemEn are input into the judgment function model. The intermediate data is selected, and the weighted average is calculated according to the kn corresponding to the intermediate data to obtain the corresponding weighted average value. The weighted average values corresponding to all kn are calculated, and the weighted average values are distributed. The values distributed in the set intervals at the head and tail are removed, and the values corresponding to the TemEn that have not been removed are output. Then the output values are converted into the environmental result temperature of the model, and the floating-point data type values are converted into temperature values, denoted as TemEf.

[0072] Set a comparison function model, input TemS and TemEf into the comparison function model to obtain the judged temperature difference Mnum value. Set the M value, where the M value represents the cooling mode, heating mode, or air supply mode of the system. The cooling mode, heating mode, or air supply mode of the system is represented as 0, 1, and 2 respectively. According to the M value, judge the judged temperature difference Mnum value to obtain the corresponding return value.

[0073] The comparison function model is represented by the following formula:

[0074] TemEf - TemS = Mnum;

[0075] In the above formula, Mnum represents the judged temperature difference Mnum value.

[0076] The specific method for judging the output value of the comparison function model to obtain the corresponding return instruction is as follows:

[0077] (1) When the M value is 0:

[0078] When Mnum ≥ 2, the corresponding return value is to adjust the cooling degree to the high - cool value high$cool.

[0079] When 0 < Mnum < 2, the corresponding return value is to adjust the cooling degree to the mid - cool value mid$cool.

[0080] When - 1 < Mnum ≤ 0, the corresponding return value is to adjust the cooling degree to the low - cool value low$cool.

[0081] When Mnum ≤ - 1, the corresponding return value is stop.

[0082] (2) When the M value is 1:

[0083] When - Mnum ≥ 2, the corresponding return value is to adjust the heat source output to the high - heat value high$heat.

[0084] When 0 < - Mnum < 2, the corresponding return value is to adjust the heat source output to the medium - heat value mid$heat.

[0085] When - 1 < - Mnum ≤ 0, the corresponding return value is to adjust the heat source output to the low - heat value low$heat.

[0086] When - Mnum ≤ - 1, the corresponding return value is stop.

[0087] (3) When the M value is 2, the corresponding return value is air.

[0088] In the parameter processing module, the pcs obtained from the data acquisition and conversion module, the TemEf obtained from the TemK (TemE, K) algorithm, and the return value are converted into integer values.

[0089] For the pcs obtained from the data acquisition and conversion module, it is converted into an octal number numP, which is specifically represented as follows:

[0090] pcs = 0numP.

[0091] For the conversion of TemEf into an octal number numT, it is specifically represented as follows:

[0092] TemEf = 0numT.

[0093] For the return value obtained from the first judgment module, the preferred conversion standard is a hexadecimal integer value encoding, which is specifically represented as follows:

[0094] stop = 0x0;

[0095] high$cool = 0x1;

[0096] mid$cool = 0x2;

[0097] low$cool = 0x3;

[0098] high$heat = 0x4;

[0099] mid$heat = 0x5;

[0100] low$heat = 0x6;

[0101] air = 0x7.

[0102] In the parameter output module, the integer values obtained from the parameter processing module are combined to form a string parameter. The string parameter is an Action string parameter, and the string parameter is represented according to the following formula:

[0103] return′ + "OCT" + pcs′ + "TEM" + teme;

[0104] In the above formula, return′ represents the integer value corresponding to the return value, "OCT" and "TEM" both represent identification characters, pcs′ represents the octal encoding value corresponding to pcs, and teme represents the octal encoding value corresponding to TemEf.

[0105] The device execution module includes an information encoding unit, a human behavior control unit, a judgment condition parameter module, a second judgment module, a control module, a physical execution unit, and a behavior message queue module.

[0106] In the human behavior control unit, the staff issues manual physical behavior instructions through operation tools such as buttons or operation methods such as remote interfaces, encodes the issued manual physical behavior instructions with Action, adds the REN identification header information before encoding and then outputs. The physical behavior instructions cover the content related to the air conditioning system, lighting system, and power supply system.

[0107] In the information encoding unit, the string parameters output by the parameter output module are used as the input of the information encoding unit. The information encoding unit first parses out the Action parameters Act_string, teme, and pcs from the string parameters according to the communication protocol, and then translates the parsed Act_string, teme, and pcs into the corresponding first execution instruction code Ac1 according to the control encoding rule, and adds the judgment identification header information J to form the first parameter JAc. The control encoding rule is selected as the hexadecimal encoding instruction set; the information encoding unit outputs the first parameter, sets the message queue list, and stores the first parameter in the message queue list;

[0108] Or, in the information encoding unit, the Action code with the REN identification header information output by the human behavior control unit is used as the input of the information encoding unit. The information encoding unit translates the Action code with the REN identification header information into the corresponding second execution instruction code Ac2 according to the control encoding rule, and adds the judgment identification header information h to form the second parameter hAc; the information encoding unit outputs the second parameter, and the second parameter will directly activate the subsequent execution flow.

[0109] In the judgment condition parameter module, human body parameters, radar parameters, and external environment comprehensive parameters are set; the human body parameters are denoted as Movh, and Movh is preferably 1, indicating "someone"; the radar parameters are denoted as Movr, and Movr is preferably 0, indicating "no object movement"; the external environment comprehensive parameters are denoted as Envp, and the external environment comprehensive parameters include temperature, humidity, sunlight, etc., and are obtained by sensor collection.

[0110] In the control module, an activation period, PCS parameters, AI parameters, and carbon consumption coefficient C are set. The activation period is the Atime parameter, and the Atime parameter is a time value, and the Atime parameter is set to 600s. The AI parameter is expressed as ai_Model(0,1,2).

[0111] In the second judgment module, the message queue list or the second parameter output by the information encoding unit, the human body parameter Movh, the radar parameter Movr, and the external environment comprehensive parameter Envp output by the judgment condition parameter module, and the AI parameter, PCS parameter, and carbon consumption parameter C output by the control module are used as the inputs of the second judgment module. There are two activation modes in the second judgment module. One is activated by the control module according to the periodic activation mode, and the periodic activation mode algorithm is adopted; the other is the direct human behavior activation mode, and the direct human behavior activation mode algorithm is adopted.

[0112] In the second judgment module, when the control module activates it periodically according to the periodic activation mode, the ai_Model parameter in the AI parameter is judged, and the first parameter in the message queue list is processed. The specific method is as follows:

[0113] First step, obtain the instruction in the message queue list and input it into the judgment function F. If the return value is null, do not proceed to the next step; otherwise, proceed to the second step.

[0114] Second step, parse teme, pcs, and Act_string1 from the returned JAc, and judge the carbon consumption change Pcs′ of the entire air conditioning system after Act_string1 is executed according to the carbon consumption parameter C. The specific judgment method is as follows:

[0115] If pcs + Pcs′ > PCS, then the intermediate parameter pcsbool = false;

[0116] If pcs + Pcs′ ≤ PCS, then the intermediate parameter pcsbool = true.

[0117] Third step, when pcsbool = true and the ai_Model parameter is 2, proceed to the fourth step; otherwise, do not proceed to the next step and stop.

[0118] Fourth step, obtain the external temperature value. When the temperature value after teme translation is greater than the external temperature value, do not execute all instructions containing heating; when teme is less than the external temperature, do not execute all instructions containing cooling. According to different situations, edit the corresponding instructions in Act_string1 to form the first execution command instruction and output it to the physical execution unit.

[0119] In the second judgment module, when it is activated according to the direct human behavior activation mode, that is, activated by an external physical behavior instruction, it is immediately activated when receiving the second parameter, and the second parameter is parsed to obtain the execution parameter Act_string 2， Output the second execution command instruction; then judge the ai_Model parameter:

[0120] When the ai_Model parameter is 0, send the second execution command instruction to the physical execution unit.

[0121] When the ai_Model parameter is 1, send the second execution command instruction to the physical execution unit, and at the same time, form the data related to the second execution command instruction into behavior data, save it as a json file, and send it to the behavior message queue module.

[0122] When the ai_Model parameter is 2, perform the following steps:

[0123] First step, call the last first parameter in the message queue list, parse the teme in it, obtain the external temperature, and then make a judgment. When the temperature value after the translation of teme is greater than the external environment temperature, do not execute all instructions containing heating. When the temperature value after the translation of teme is less than the external temperature, do not execute all instructions containing cooling; in other cases, perform the second step.

[0124] Second step, obtain the instructions in the message queue list, input them into the judgment function F. If the return value is null, do not proceed to the next step. Otherwise, proceed to the third step.

[0125] Third step, parse teme, pcs, and Act_string1 from the returned first parameter, and judge the carbon consumption change Pcs′ of the entire air conditioning system after the execution of Act_string1 according to the carbon consumption parameter C. The specific judgment method is:

[0126] If pcs + Pcs′ > PCS, stop;

[0127] If pcs + Pcs′ ≤ PCS, trigger the execution command and edit the relevant execution instructions in Act_string1.

[0128] Fourth step, pass Act_string1 to the physical execution unit, and at the same time, form the data related to the execution instruction into behavior data, save it as a json file, and send it to the behavior message queue module.

[0129] The physical execution unit receives the first and second execution command instructions output by the second judgment module, parses the first and second execution command instructions, decomposes them into an instruction set queue, and mobilizes the corresponding devices to execute the queue for the instruction set.

[0130] The behavior message queue module includes a behavior message queue AC_H_MQ composed of multiple behavior data in JSON file format. This behavior message queue is a collection of behavior data directly controlled by humans within a time period, and all include control user information, control terminal information, control mode, electrical state, control instruction content, and control relative time. Among them, the control instruction content is parsed from the Ac part, the control relative time is passed by the Timer parameter, and the remaining data is parsed from the header file R of hAc. The behavior message queue is a list of JSON data.

[0131] The carbon behavior deep learning module includes a deep learning setting module, a behavior learning module, a refresh module, and a control information setting module.

[0132] In the deep learning setting module, the activation period T of deep learning and the activation timer Timer after startup activation are set; the PCS, TemS, Atime, sensitivity parameter K, and ai_Model parameter in deep learning are also set. All settings in the deep learning setting module can be set manually or according to the large model. When the ai_Model parameter of the set deep learning is 0, the behavior learning module is not activated for learning. Otherwise, the behavior learning module is activated for learning.

[0133] The behavior learning module includes a first AI model, a second AI model, and an AI behavior tree. The first AI model is a deep learning model, and the second AI model is a reinforcement learning model. The first AI model and the second AI model operate independently of each other, are responsible for learning and evolving for behavior data, and give independent parameter adjustment results. The AI behavior tree is responsible for selecting between the results of the two models and adjusting the probability ratio between the two selections.

[0134] The input of the behavior learning module is behavior data, which is the processed behavior message queue. The processed behavior message queue is split into multiple information lists, and each information list only includes one kind of split information.

[0135] The structure of the first AI model consists of three layers. The first layer is a linear layer responsible for processing the input data. The second layer is a non - linear layer that uses the tanh function as the basic activation function and the Swish function as the function to accelerate learning to adapt to the situation where the number of samples of the behavior data itself is relatively small. The third layer is an output layer, which is a combination of a linear function and a sigmoid function. In the first layer, the transformer method is used to process multiple time - value tags Timer in the input behavior data into interval parameter Atime′, process TemS in the behavior data into TemS′, and process pcs in the behavior data into carbon consumption parameter PCS′. Then, Atime′, TemS′, PCS′, N, and TemS are input into the deep - learning model, where the parameter N represents the number and frequency of human interventions. The deep - learning model outputs Atime parameter, PCS parameter, TemS parameter, and K parameter, and stores the output parameters in the deep - learning database.

[0136] In the second AI function, TemS in the behavior data is processed into TemS′, and TemS′ is compared with TemS. According to the Qlearning method, the correction value list of the output TemS parameter is adjusted, and the output probability ratio of each value in the correction value list of the TemS parameter is adjusted. The number of time markers Timer in the behavior data determines the number of times the reinforcement learning model is activated in a learning cycle. When the last learning is completed, the reinforcement learning model will output the TemS parameter result, and based on the difference between the TemS parameter result and the TemS setting result of the previous cycle, the corresponding Atime parameter, PCS parameter, and K parameter are given.

[0137] At the end of a learning cycle, the AI behavior tree calls the results of the deep - learning model and the reinforcement learning model, and uses the Qlearning method to select the results of the two according to the weight parameter E. The weight parameter E is corrected according to the adjustment times parameter N of the previous cycle, and finally the recommended PCS value and TemS value, as well as the control interval parameters Atime and K parameter are given and updated.

[0138] The input behavior data, deep - learning input parameters and output results, reinforcement - learning input parameters and output results, and AI behavior tree weight parameter E are stored in an AI learning behavior JSON file, and the file is stored in the AI deep - learning library after adding the system timestamp.

[0139] When the ai_Model parameter of the deep - learning is equal to 1, information is sent to the refresh module. When the ai_Model parameter of the deep - learning is equal to 2, the result is sent to the control information setting module.

[0140] After receiving the information, the refresh module clears all instruction information in the behavior message queue.

[0141] After receiving the information, the control information setting module converts the TemS and K parameters into strings, encodes them to form a first encoded file, and transports the first encoded file to the environmental carbon judgment module for update; converts the Atime and PCS into strings, encodes them to form a second encoded file, and transports the second encoded file to the device execution module for update; and activates the refresh module simultaneously.

[0142] The present invention sets up three parts of modules. Each of the three parts of modules has a complete basic artificial intelligence module and an interaction module and can be applied independently. Therefore, they can be arranged on different terminal devices respectively, making the algorithm have the characteristic of being distributed in the physical space. After the information is connected, the three independent algorithms can be fused to jointly form a deep learning algorithm for low-carbon artificial intelligence behavior, forming a complete algorithm structure. At this time, the overall algorithm will be executed sequentially between different devices according to a specific algorithm process, and a complete behavior deep learning algorithm is realized by combining the function modules between different devices. This distributed layout enables the algorithm to penetrate into various control terminals, can fully expand the coverage range of the algorithm, and greatly enhances the adaptability and system security of the algorithm. The algorithm no longer depends on a specific device or a specific system structure to be realized. The algorithm consists of many independent parts, which means that the absence of any part will not affect the basic operation of the terminal device and the artificial intelligence function. It realizes reducing the consumption of computing resources, increasing the coverage range of artificial intelligence, and improving the adaptability and effect of artificial intelligence in actual applications in different environments.

[0143] After measurement, compared with the current algorithm arranged on the server or the edge computing core, the consumption of computing resources of this algorithm can be reduced by more than 50%, and the coverage range of artificial intelligence can be increased by more than 40%.

[0144] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than a limitation on the protection scope of the present invention. Any simple modification or equivalent replacement made by those of ordinary skill in the art to the technical solution of the present invention does not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A building environment carbon reduction system based on artificial intelligence behavior deep learning, characterized in that: It includes an environmental carbon judgment module, a device execution module and a carbon behavior deep learning module. The input of the environmental carbon judgment module is the first data. The environmental carbon judgment module converts the input first data into a string parameter and outputs it. The input of the device execution module is the string parameter output by the environmental carbon judgment module. The device execution module processes the input string parameter and then outputs a control instruction and a behavior message queue. The input of the carbon behavior deep learning module is the behavior message queue. The carbon behavior deep learning module processes the behavior message queue to obtain a first encoding file and a second encoding file. The environmental carbon judgment module is updated according to the first encoding file, and the device execution module is updated according to the second encoding file.

2. According to claim 1, a building environment carbon reduction system based on artificial intelligence behavior deep learning is characterized in that: The environmental carbon judgment module includes a data collection and conversion module, a judgment parameter input and conversion module, a first judgment module, and a parameter output module; The data acquisition and conversion module is used to obtain the first data, which includes the ambient temperature TemE and the power of the power system. The data acquisition and conversion module removes the green energy part in the power of the power system to obtain the carbon consumption power: the judgment parameter input and conversion module is used to construct a time data list, obtain the set temperature TemS, set time and sensitive parameter K; the first judgment module is used to judge the input carbon consumption power, ambient temperature, set temperature, set time and sensitive parameters to obtain a return value; the parameter output module is used to form a string parameter according to the return value obtained by processing the first judgment module.

3. The building environment carbon reduction system based on artificial intelligence behavior deep learning according to claim 2 is characterized in that: In the first judgment module, the specific method for obtaining the return value is: The input data is corrected using the TemK (TemE, K) algorithm. The data is first screened and a constant k is set. According to the set constant k, adjacent data whose absolute value of the ambient temperature change is less than |k| are merged to obtain a merged data set. The merged data set includes multiple merged data pairs (TemEn, kn), where TemEn represents the nth ambient temperature value in the merged data set, and kn represents the time value corresponding to TemEn. A judgment function model is set, and the function variables are TemEn and kn. The sigmoid function is used as the activation function in the judgment function model. The ambient temperature in the merged data set is judged by the function, and the ambient result temperature after judgment is recorded as TemEf. A comparison function model is set, and the set temperature TemS in the first data and the ambient result temperature TemEf processed by the judgment function model are input into the comparison function model to calculate the corresponding judgment temperature difference Mnum value. The judgment temperature difference Mnum value is calculated specifically according to the following formula: TemEf-TemS=Mnum; In the above formula, Mnum represents the determined temperature difference, and TemS represents the set temperature in the first data; Input Mnum and the corresponding M value into the judgment part of the comparison function, where the M value is set to 0, 1, or 2. The specific method of obtaining the return value through the judgment part of the comparison function model is: (1) When the M value is 0: When Mnum ≥ 2, the corresponding return value is to adjust the refrigeration degree to the high - cool value high$cool; When 0 < Mnum < 2, the corresponding return value is to adjust the refrigeration degree to the mid - cool value mid$cool; When - 1 < Mnum ≤ 0, the corresponding return value is to adjust the refrigeration degree to the low - cool value low$cool; When Mnum ≤ - 1, the corresponding return value is stop; (2) When the M value is 1: When - Mnum ≥ 2, the corresponding return value is to adjust the heat source output to the high - heat value high$heat; When 0 < - Mnum < 2, the corresponding return value is to adjust the heat source output to the medium - heat value mid$heat; When - 1 < - Mnum ≤ 0, the corresponding return value is to adjust the heat source output to the low - heat value low$heat; When - Mnum ≤ - 1, the corresponding return value is stop; (3) When the M value is 2, the corresponding return value is air; The string parameter formed by the parameter output module is represented by the following formula: return′ + "OCT” + pcs′ + "TEM” + teme; In the above formula, return′ represents the integer value corresponding to the return value, "OCT” and "TEM” both represent identification characters, pcs′ represents the octal encoding value corresponding to pcs, pcs represents the carbon consumption power, and teme represents the octal encoding value corresponding to TemEf.

4. The building environment carbon reduction system based on artificial intelligence behavior deep learning according to claim 3 is characterized in that: The device execution module includes an information encoding unit, a judgment condition parameter module, a second judgment module, a control module, a physical execution unit, and a behavior message queue module; The information encoding unit inputs the string parameter, and the information encoding unit processes the string parameter to obtain a first parameter, which is a first execution instruction code with a judgment identification header information J; The first parameter is stored in the message queue list, and the information encoding unit outputs the message queue list; or, the information encoding unit inputs an Action code with a REN identification header information, and the information encoding unit processes the Action code with a REN identification header information to obtain a second parameter, which includes a second execution instruction code with a judgment identification header information h; The judgment condition parameter module is used to set human body parameters, radar parameters, and external environment comprehensive parameters; the control module is used to set the activation period, PCS parameters, AI parameters, and carbon consumption coefficient; the second judgment module inputs human body parameters, radar parameters, external environment comprehensive parameters, AI parameters, PCS parameters, and carbon consumption parameters; The control module activates the second judgment module according to the activation period, and the second judgment module processes the information in the message queue list and outputs a first execution command instruction; or, the second judgment module is directly activated according to the received physical behavior instruction, and the second judgment module is immediately activated after receiving the second parameter and outputs a second execution command instruction and a behavior data json file; The physical execution unit is used to parse the first and second execution command instructions and decompose them into instruction set queues, and mobilize devices to perform queue execution according to different instruction sets; the behavior message queue module includes multiple json file format behavior data, and the behavior message queue is a collection list of json file format behavior data for a time period.

5. The building environment carbon reduction system based on artificial intelligence behavior deep learning according to claim 4 is characterized in that: The control module activates the second judgment module according to the control cycle. When activated, the ai_Model parameter in the AI ​​parameter is judged and the first parameter in the message queue list is processed. The specific method is: The first step is to obtain the instructions in the message queue list and input them into the judgment function F. If the return value is null, the next step will not be performed. Otherwise, the second step will be performed. The second step is to parse out teme, pcs and Act_string1 from the first parameter returned, and determine the carbon consumption change Pcs′ of the entire air conditioning system after Act_string1 is executed based on the carbon consumption parameter. The specific determination method is: pcs+Pcs′>PCS, then the intermediate parameter pcsbool=false; pcs+Pcs′≤PCS, then the intermediate parameter pcsbool=true; Step 3: When pcsbool=true and the ai_Model parameter is 2, proceed to step 4. Otherwise, do not proceed to the next step and stop. The fourth step is to obtain the external temperature value. When the temperature value after teme translation is greater than the external temperature value, all instructions containing heating will not be executed. When the temperature value after teme translation is less than the external temperature value, all instructions containing cooling will not be executed. According to different situations, the corresponding instructions are edited in Act_string1 to form an execution command instruction and output to the physical execution unit.

6. The building environment carbon reduction system based on artificial intelligence behavior deep learning according to claim 5 is characterized in that: In the second judgment module, it is directly activated according to the behavior. It is activated upon receiving the second parameter, and the second parameter is parsed to obtain the execution parameter Act_string2. Then the ai_Model parameter is judged: When the ai_Model parameter is 0, the second execution command is sent to the physical execution unit; When the ai_Model parameter is 1, the second execution command instruction is sent to the physical execution unit, and the second execution command instruction related data is combined into behavior data, saved as a json file, and sent to the behavior message queue module; When the ai_Model parameter is 2, the following steps are performed: The first step is to call the last first parameter in the message queue list and parse the teme in it to obtain the external environment temperature, and then make a judgment. When the temperature value after teme is translated is greater than the external environment temperature value, all instructions containing heating will not be executed. When the temperature value after teme is translated is less than the external environment temperature value, all instructions containing cooling will not be executed. In other cases, proceed to the second step; The second step is to obtain the instruction in the message queue list and input it into the judgment function F. If the return value is null, the next step will not be performed. Otherwise, the third step will be performed. The third step is to parse out teme, pcs and Act_string1 from the first parameter returned, and determine the carbon consumption change Pcs′ of the entire air conditioning system after Act_string1 is executed according to the carbon consumption parameter C. The specific determination method is: pcs+Pcs′>PCS, then stop; If pcs+Pcs′≤PCS, the execution command is triggered and the relevant execution instructions are edited in Act_string1; The fourth step is to pass Act_string1 to the physical execution unit, and at the same time, the execution instruction related data is combined into behavior data, saved as a json file, and sent to the behavior message queue module.

7. The building environment carbon reduction system based on artificial intelligence behavior deep learning according to claim 1 is characterized in that: The carbon behavior deep learning module includes a deep learning setting module, a behavior learning module, a refresh module, and a control information setting module; The deep learning setting module is provided with an ai_Model parameter. Only when the ai_Model parameter is 0, the behavior learning module is not activated, otherwise the behavior learning module will be activated; the behavior learning module includes a first AI model, a second AI model and an AI behavior tree. The first AI model and the second AI model are both subjected to learning evolution for the behavior data, and corresponding parameter adjustment results are given. The AI ​​behavior tree is responsible for selecting one of the models, adjusting the probability ratio between the two selections, and outputting the PCS value, TemS value, Atime parameter and K parameter; when the ai_Model parameter is 1, the output value is sent to the refresh module, and when the ai_Model parameter is 2, the output value is sent to the control information setting module; when the refresh module receives the output value, all instruction information in the behavior message queue is cleared; when the control information setting module receives the information, the TemS value and the K parameter are converted into a character string, and encoded to form a first encoding file, and the first encoding file is transmitted to the environmental carbon judgment module for updating; the Atime parameter and the PCS value are converted into a character string, and encoded to form a second encoding file, and the second encoding file is transmitted to the device execution module for updating; At the same time, the refresh module is activated to clear all instruction information in the behavior message queue.

8. The building environment carbon reduction system based on artificial intelligence behavior deep learning according to claim 7 is characterized in that: The first AI model is a deep learning model, including a linear layer, a nonlinear layer and an output layer, the linear layer is used to input behavior data, the linear layer uses the tansformer method to obtain Atime parameters, PCS parameters, TemS parameters and K parameters, and the output layer is used to output Atime parameters, PCS parameters, TemS parameters and K parameters; The second AI model is a reinforcement learning model, which uses the Qlearning method to obtain the output Atime parameters, PCS parameters and K parameters.

9. The building environment carbon reduction system based on artificial intelligence behavior deep learning according to claim 8 is characterized in that: The AI ​​behavior tree uses the Qlearning method to select the output results of the first AI model and the second AI model according to the weight parameter E, correct the weight parameter E according to the adjustment number parameter of the previous cycle, and output the recommended PCS value, TemS value, Atim parameter and K parameter.

Images