WO2021082511A1 - 模型训练方法、控制参数确定方法及装置 - Google Patents

模型训练方法、控制参数确定方法及装置 Download PDF

Info

Publication number
WO2021082511A1
WO2021082511A1 PCT/CN2020/100520 CN2020100520W WO2021082511A1 WO 2021082511 A1 WO2021082511 A1 WO 2021082511A1 CN 2020100520 W CN2020100520 W CN 2020100520W WO 2021082511 A1 WO2021082511 A1 WO 2021082511A1
Authority
WO
WIPO (PCT)
Prior art keywords
energy consumption
offset
control
current
conditioning system
Prior art date
Application number
PCT/CN2020/100520
Other languages
English (en)
French (fr)
Inventor
袁德玉
汤潮
Original Assignee
北京国双科技有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京国双科技有限公司 filed Critical 北京国双科技有限公司
Publication of WO2021082511A1 publication Critical patent/WO2021082511A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F24—HEATING; RANGES; VENTILATING
    • F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
    • F24F11/00—Control or safety arrangements
    • F24F11/62—Control or safety arrangements characterised by the type of control or by internal processing, e.g. using fuzzy logic, adaptive control or estimation of values
    • F24F11/63—Electronic processing
    • F24F11/64—Electronic processing using pre-stored data
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F24—HEATING; RANGES; VENTILATING
    • F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
    • F24F11/00—Control or safety arrangements
    • F24F11/88—Electrical aspects, e.g. circuits
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Definitions

  • This application relates to the technical field of energy saving and consumption reduction, in particular to a model training method, a method and device for determining control parameters.
  • the energy saving and consumption reduction of the air conditioning system is an important part of the field of energy saving and consumption reduction.
  • the existing energy consumption reduction technology for the air conditioning system uses an energy consumption prediction model to obtain the control parameters at the lowest energy consumption output by the energy consumption prediction model.
  • control parameters obtained by the prior art are the control parameters when the power consumption output by the energy consumption prediction model is the lowest.
  • the control parameters obtained by the energy consumption prediction model in the prior art are applied to the air conditioning system, the effect of effectively reducing the energy consumption of the air conditioning system is often not achieved.
  • this application provides a model training method, a control parameter determination method and a device that overcome the above problems or at least partially solve the above problems.
  • the solutions are as follows:
  • a model training method including:
  • the training data is obtained according to the test records, and the current offset determination model is trained using the training data, where the training data includes: a second environmental parameter group and an actual offset.
  • each test record contains at least a second environmental parameter group, an actual offset, and energy consumption parameters.
  • the training data is obtained according to the test record, and the training data is used to compare the current
  • the offset determination model for training includes:
  • the obtained actual offset is used as the expected output of the current offset determination model, and the current offset determination model is trained, wherein the input corresponding to the expected output is: The second environmental parameter group corresponding to the actual offset.
  • the energy consumption parameter is an energy consumption ratio
  • each test record contains at least a second environmental parameter group, a control parameter, an actual offset, and an energy consumption parameter.
  • the energy consumption ratio Actual energy consumption of the air-conditioning system/reference energy consumption of the air-conditioning system.
  • the actual offset is used as a control offset
  • the current second environmental parameter group, the control parameter, the actual offset, and the energy consumption parameter are used as training data
  • the current energy consumption Training is performed on the ratio determination model, wherein the input of the energy consumption ratio determination model is the second environmental parameter group, the control parameter and the control offset, and the output of the energy consumption ratio determination model is the energy consumption ratio.
  • the obtaining training data according to test records and using the training data to train the current offset determination model includes:
  • the current energy consumption ratio is determined to be the minimum energy consumption ratio output by the model The control offset
  • the actual offset that minimizes the energy consumption ratio output by the current energy consumption ratio determination model is used as the expected output of the current offset determination model, and the current offset determination model is trained, where
  • the input corresponding to the expected output is: a second environmental parameter group corresponding to the expected output in the test record.
  • the obtaining training data according to test records and using the training data to train the current offset determination model includes:
  • the output of the current offset determination model is used as one of the inputs of the current energy consumption ratio determination model, and the current energy consumption ratio determination model outputs the minimum energy consumption ratio as the goal, obtained from test records Training data, using the training data to train the current offset determination model.
  • the energy consumption parameter is an energy consumption ratio
  • the obtaining the energy consumption parameter of the air conditioning system when the air conditioning system is operating under the test control parameter includes:
  • energy consumption ratio actual energy consumption/reference energy consumption of the air conditioning system, the energy consumption ratio is calculated.
  • a method for determining control parameters includes:
  • a model training device includes: a first input unit, a second input unit, a disturbance generating unit, a test parameter obtaining unit, a test record generating unit, and a first training unit,
  • the first input unit is configured to input the current first environmental parameter group of the air-conditioning system into the energy consumption prediction model obtained by training, and obtain the control parameters when the energy consumption output by the energy consumption prediction model is the lowest.
  • the input of the energy consumption prediction model is the first environmental parameter group and the control parameter, and the output of the energy consumption prediction model is the energy consumption of the air conditioning system;
  • the second input unit is configured to input the current second environmental parameter group into the current offset determination model to obtain the control offset output by the current offset determination model, wherein the offset
  • the input of the quantity determination model is the second environmental parameter group, and the output is the control offset
  • the disturbance amount generating unit is configured to generate a disturbance amount, and determine an actual offset according to the disturbance amount and the control offset;
  • the test parameter obtaining unit is configured to obtain a set of test control parameters of the air conditioning system according to the actual offset and the obtained control parameter;
  • the test record generating unit is configured to obtain the energy consumption parameters of the air-conditioning system when operating under the set of control parameters for the test, and generate a test record;
  • the first training unit is configured to obtain training data according to test records, and use the training data to train the current offset determination model, wherein the training data includes: a second environmental parameter group and actual offset Shift.
  • a control parameter determination device includes: a control parameter obtaining unit, an offset obtaining unit, and a target parameter determining unit,
  • the control parameter obtaining unit is configured to input the current first environmental parameter group of the air-conditioning system into the energy consumption prediction model obtained by training, and obtain the control parameter when the energy consumption output by the energy consumption prediction model is the lowest.
  • the input of the energy consumption prediction model is the first environmental parameter group and the control parameter, and the output of the energy consumption prediction model is the energy consumption of the air conditioning system;
  • the offset obtaining unit is configured to input the current second environmental parameter group into the current offset determination model to obtain the control offset output by the current offset determination model, wherein the offset
  • the input of the displacement determination model is the second environmental parameter group, and the output is the control offset
  • the target parameter determining unit is configured to determine the target control parameter of the air conditioning system according to the obtained control parameter and the obtained control offset.
  • a storage medium includes a stored program, wherein the device where the storage medium is located is controlled to execute any of the above-mentioned model training methods and/or the above-mentioned control parameter determination method when the program is running.
  • a device comprising at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other through the bus; the processing The device is used to call the program instructions in the memory to execute any of the above-mentioned model training methods and/or the above-mentioned control parameter determination methods.
  • the model training method, control parameter determination method and device can obtain the control parameters at the lowest energy consumption according to the energy consumption prediction model, and determine the model to obtain the control bias according to the current offset.
  • the actual offset is determined according to the disturbance and the control offset; according to the actual offset and control parameters, a set of test control parameters for the air-conditioning system is obtained; the air-conditioning system is obtained under the set of test control parameters Generate a test record based on the energy consumption parameters at runtime; obtain training data according to the test records, and use the training data to train the current offset determination model.
  • the model training method of the present application can be trained to obtain the offset determination model, and the control offset of the model output is determined by the offset, which can be superimposed on the energy consumption prediction model to obtain the control parameters when the energy consumption is the lowest. , So as to obtain the target control parameters that can truly reduce the energy consumption of the air conditioning system.
  • this application uses the addition of disturbance before training to enrich the test data and improve the accuracy and applicability of the offset determination model.
  • Fig. 1 shows a flowchart of a model training method provided by an embodiment of the present application
  • Figure 2 shows a flowchart of another model training method provided by an embodiment of the present application
  • FIG. 3 shows a flowchart of a method for determining control parameters provided by an embodiment of the present application
  • FIG. 4 shows a schematic structural diagram of a model training device provided by an embodiment of the present application
  • FIG. 5 shows a schematic structural diagram of a control parameter determination device provided by an embodiment of the present application
  • Fig. 6 shows a schematic structural diagram of a device provided by an embodiment of the present application.
  • the energy consumption prediction model is a machine model obtained by training the collected historical data as training data, where the historical data includes: the corresponding first environmental parameter group, control parameters, and energy consumption of the air-conditioning system.
  • the first environmental parameter group may include: the required refrigeration capacity of the air-conditioning system and/or the uncontrolled temperature of the air-conditioning system, etc.
  • the required refrigeration capacity of the air-conditioning system can be determined according to environmental parameters such as indoor and outdoor temperature and humidity.
  • the uncontrolled temperature of the air-conditioning system may include: evaporating temperature, condensing temperature, uncontrolled temperature during the operation of the air-conditioning system, etc.
  • control parameters obtained through the energy consumption prediction model often not the control parameters that can effectively reduce the energy consumption of the air conditioning system. It is also often not the optimal control parameter that can minimize the energy consumption of the air-conditioning system.
  • the structural design of the energy consumption prediction model itself may also cause the control parameters obtained through the energy consumption prediction model to often not be the control parameters that can effectively reduce the energy consumption of the air conditioning system.
  • the existing technicians generally strive to improve the accuracy of the energy consumption prediction model by improving the energy consumption prediction model. For example: collecting more historical data, optimizing the structure of energy consumption prediction models, etc.
  • the research direction of the inventor of this application is based on the control deviation of the control parameters at the lowest energy consumption obtained according to the energy consumption prediction model and the optimal control parameters at the lowest actual energy consumption of the air-conditioning system. Amount.
  • the inventor of the present application proposes an offset determination model to determine the control offset.
  • the input of the offset determination model is the second environmental parameter group, and the output is the control offset.
  • control parameters at the lowest energy consumption are obtained according to the energy consumption prediction model, only the control offset is determined according to the offset determination model of the application, and the control parameters and the control offset can be obtained according to the above control parameters and control offset.
  • the target control parameter of the energy consumption of the air conditioning system is determined according to the offset determination model of the application, and the control parameters and the control offset can be obtained according to the above control parameters and control offset.
  • this application first provides a model training method to train the offset determination model.
  • the application can obtain the offset determination model, and can also continuously improve the offset determination model.
  • an embodiment of the present application provides a model training method, which may include:
  • the required refrigeration capacity of the air conditioning system can be determined according to environmental parameters such as indoor and outdoor temperature and humidity
  • the first environmental parameter group can also include: outdoor temperature, outdoor humidity, thermodynamics At least one of environmental parameters such as wet bulb temperature, wind direction, wind speed, indoor temperature, indoor maximum temperature, indoor humidity, etc.
  • the control parameters may include at least one of the inlet water temperature of the air-conditioning system, the outlet water temperature of the air-conditioning system, the frequency of the water pump, and the flow rate of cold water.
  • energy consumption refers to the electrical energy consumed by the operation of the air-conditioning system, which can be determined according to the power of the air-conditioning system.
  • the second environmental parameter group may include outdoor environmental parameters and indoor environmental parameters, where the outdoor environmental parameters may include at least one of outdoor temperature, outdoor humidity, thermodynamic wet bulb temperature, wind direction, wind speed and other parameters.
  • indoor environmental parameters may include: indoor temperature, indoor maximum temperature, indoor humidity, and so on.
  • the indoor environment parameters may also include: the power of the indoor electronic devices, etc.
  • the second environmental parameter group and the first environmental parameter group may or may not have an intersection.
  • the second environmental parameter group may be a proper subset of the first environmental parameter group.
  • step S100 and step S200 are not limited in this application.
  • the second environmental parameter group and the control parameter may respectively include multiple parameters
  • a vector may be used in this application to represent the second environmental parameter group and the control parameter.
  • the second environmental parameter group is [35, 90, 25, 30, 70]
  • this one-dimensional vector represents: outdoor temperature 35 degrees, outdoor humidity 90%, indoor temperature 25 degrees, indoor maximum temperature 30 degrees, indoor humidity 70 %.
  • multidimensional vectors can also be used to represent the second environmental parameter group and control parameters, which are not limited in this application.
  • the control parameter is [12, 17, 40, 558]
  • the one-dimensional vector represents: the inlet water temperature is 12 degrees, the outlet water temperature is 17, the pump frequency is 40 Hz, and the cold water flow rate is 558 cubic meters per hour.
  • the current offset determination model is the initial offset determination model.
  • the initial offset determines that the control offset of the model output is all 0.
  • control offset of the model's output will generally not be all 0 again.
  • the structure of the vector of the control offset output by the offset determination model and the structure of the vector of the control parameter obtained according to the energy consumption prediction model may be the same, for example, both may have a 4 ⁇ 4 structure.
  • the meaning of each element in the vector of control offset output by the offset determination model is the same as the meaning of the element at the same position in the vector of control parameters obtained from the energy consumption prediction model.
  • the control offset obtained in step S200 may be [+0.5, -0.3, +4, -8], and this one-dimensional vector represents: The temperature offset is +0.5 degrees, the outlet water temperature offset is -0.3 degrees, the pump frequency offset is +4 Hz, and the cold water flow offset is -8 cubic meters per hour.
  • control offset is not only related to the second environmental parameter group, but also related to the control parameters obtained according to the energy consumption prediction model.
  • the energy consumption prediction model obtains the control parameters indirectly based on the second environmental parameter group.
  • the input of the offset determination model in this application is only the second environmental parameter group.
  • the present application can generate the disturbance quantity in a variety of ways, for example, a random number is generated by a random number generation algorithm, and the random number is used as the disturbance quantity.
  • a random number is generated by a random number generation algorithm, and the random number is used as the disturbance quantity.
  • the actual offset determined in step S300 is only a hypothetical actual offset, and does not represent the real actual offset.
  • the disturbance can also be a vector
  • the offset determines the structure of the vector of the control offset output by the model and the structure of the vector of the disturbance can be the same.
  • the offset determines the vector of the control offset output by the model.
  • the meaning of each element is the same as the meaning of the element at the same position in the disturbance vector.
  • the disturbance generated in step S300 can be: [+0.1, -0.1, +0.4, -2 ]
  • the one-dimensional vector represents: the disturbance of the inlet water temperature is +0.1 degree
  • the disturbance of the outlet water temperature is -0.1 degree
  • the disturbance of the pump frequency is +0.4 Hz
  • the disturbance of the cold water flow is -2 cubic meters /per hour.
  • the amount of disturbance can also be expressed in proportion.
  • the amount of disturbance generated in step S300 can be: [+5%, -5%, +8%, -2%], this one-dimensional vector represents: the disturbance of the inlet water temperature
  • the amount of disturbance is +5%
  • the disturbance amount of the outlet water temperature is -5%
  • the disturbance amount of the frequency of the water pump is +8%
  • the disturbance amount of the cold water flow is -2%.
  • the disturbance quantity can be added to the control offset to obtain the actual offset. For example: add [+0.5, -0.3, +4, -8] and [+0.1, -0.1, +0.4, -2] to obtain the actual offset as [+0.6, -0.4, +4.4,- 10].
  • the corresponding proportion can be increased or decreased in the control offset according to the ratio multiplication. For example: on the basis of [+0.5, -0.3, +4, -8], float according to [+5%, -5%, +8%, -2%] to obtain [+0.525, -0.285, +4.32 , -7.84].
  • the disturbance amount can be within a preset numerical range or proportional range, for example: the disturbance amount of the inlet water temperature is in the range of ⁇ 1 degree, and the disturbance amount of the cold water flow is in the range of ⁇ 50 cubic meters per hour. .
  • the application can obtain more test data and add randomness to the test data.
  • the disturbance amount is within the preset value range, it will not cause the air-conditioning system to operate in an obviously unreasonable situation, and will not bring too much operating burden to the air-conditioning system.
  • a smaller amount of disturbance also provides a smaller test granularity, so that the coverage of the test data becomes higher, and thus the accuracy of the offset determination model obtained by training is higher.
  • the present application may determine the disturbance amount corresponding to the current second environmental parameter group in the previously generated disturbance amount as the disturbance amount to be compared, and the disturbance amount to be compared is respectively compared with The amount of disturbance generated this time is compared to determine whether the test of the amount of disturbance generated this time has been performed under the current second environmental parameter group. If there is at least one disturbance to be compared that is equal to the disturbance generated this time, the disturbance generated this time is discarded, a new disturbance is generated again and the above comparison is continued. If the disturbances to be compared are not equal to the disturbances generated this time, the actual offsets can be determined according to the disturbances generated this time and the control offsets and the subsequent steps are executed.
  • S400 Obtain a set of test control parameters of the air conditioning system according to the actual offset and the obtained control parameters.
  • the present application may add the actual offset to the obtained control parameter to obtain a set of test control parameters for the air conditioning system. For example: when the actual offset is [+0.6, -0.4, +4.4, -10] and the control parameter is [12, 17, 40, 558], superimpose the two to obtain a set of test control parameters as [12.6 , 16.6, 44.4, 548].
  • S500 Obtain energy consumption parameters of the air-conditioning system when operating under the set of control parameters for the test, and generate a test record.
  • each of the test records correspondingly stores: at least one of the second environmental parameter group, the control parameter, the control offset, the disturbance amount, and the energy consumption parameter.
  • this application can control the control parameters of the air conditioning system as the test control parameters, for example: when the test control parameters are [12.6, 16.6, 44.4, 548], the control parameters of the air conditioning system The inlet water temperature is 12.6 degrees, the outlet water temperature is 16.6 degrees, the frequency of the water pump is 44.4 Hz, and the cold water flow rate is 548 cubic meters per hour.
  • control parameters for controlling the air-conditioning system are the control parameters for this group of tests, you can wait for a period of time before collecting energy consumption parameters. This is because after the control parameters change, the air-conditioning system needs a period of time to enter a relatively stable state, and the power consumption parameters of the relatively stable state have greater practical significance.
  • this application can use a form to record each test record, and the form can be as shown in Table 1:
  • the actual energy consumption of the air-conditioning system can be determined through actual measurement.
  • the reference energy consumption of the air conditioning system output by the environmental reference energy consumption prediction model obtained by the training may be obtained by inputting the current second environmental parameter group into the environmental reference energy consumption prediction model obtained by training.
  • the environmental reference energy consumption prediction model is a machine model trained through training data (the training data includes the corresponding second environmental parameter group and the reference energy consumption of the air-conditioning system). Its input is the second environmental parameter group and the output is the air-conditioning system. System baseline energy consumption.
  • step S500 may specifically include:
  • energy consumption ratio actual energy consumption/reference energy consumption of the air conditioning system, the energy consumption ratio is calculated.
  • this application can generate multiple different disturbances, and add these disturbances to the current offset determination model according to the control obtained by the second environmental parameter group.
  • offset a number of different actual offsets are obtained.
  • the present application adds these actual offsets to the control parameters at the lowest energy consumption output by the energy consumption prediction model according to the second environmental parameter group, so as to obtain multiple sets of control parameters for testing.
  • These multiple sets of test control parameters are all test control parameters under the same second environmental parameter group.
  • This application can control the air conditioning system to operate according to the above-mentioned multiple sets of test control parameters in different time periods, so that multiple test records can be collected.
  • the second environmental parameter group, control parameters, and control offset are all the same, but the disturbances are different.
  • the present application can test the energy consumption parameters caused by multiple different disturbances in the same environment, resulting in a higher data coverage effect.
  • the present application can train the current offset determination model according to the test record in various situations.
  • the above-mentioned multiple situations can include:
  • Case 1 The number of test records is less than the first preset number.
  • test records are still relatively small, and the test needs to be continued to accumulate data.
  • the first situation may be specifically: the number of test records in the preset test time period is less than the first preset number.
  • Case 2 The current second environmental parameter group exceeds the coverage range of the second environmental parameter group in the obtained test record.
  • the lowest outdoor temperature in the second environmental parameter group in the obtained test record is minus 5 degrees
  • the current outdoor temperature in the second environmental parameter group is minus 7 degrees
  • the current second environmental parameter group exceeds The coverage of the second environmental parameter group in the obtained test record.
  • this application can effectively collect test data in more environments, so as to increase the coverage of test data and improve the accuracy and applicability of the machine model.
  • the preset first energy consumption parameter may be: the preset energy consumption ratio.
  • the preset energy consumption ratio When a certain energy consumption ratio is higher than the preset energy consumption ratio, it means that the energy consumption is relatively high and cannot achieve a better effect of reducing energy consumption. . Need to continue testing.
  • Case 4 The energy consumption parameter corresponding to the current second environmental parameter group in the obtained test record is higher than the preset first energy consumption parameter, and the test record corresponding to the current second environmental parameter group has been obtained The number of bars is less than the second preset number, wherein the second preset number is greater than the first preset number.
  • the energy consumption parameter corresponding to the current second environmental parameter group in the obtained test record is higher than the preset first energy consumption parameter, and the number of test records is less than the second preset number, where , The second preset number is greater than the first preset number.
  • each of the test records correspondingly stores at least: the second environmental parameter group, the actual offset, and the energy consumption parameter.
  • Step S600 may specifically include:
  • the obtained actual offset is used as the expected output of the current offset determination model, and the current offset determination model is trained, wherein the input corresponding to the expected output is: The second environmental parameter group corresponding to the actual offset.
  • the application can obtain multiple test records through the test after adding the disturbance, but not all the energy consumption parameters in the test records are good.
  • this application can filter the test records, filter out the test records that meet the preset energy consumption requirements, and then combine the second environmental parameter group and the actual test records in the selected test records.
  • the offset is used as training data to train the current offset determination model.
  • the aforementioned preset energy consumption requirement may be that the energy consumption parameter is not higher than the preset second energy consumption parameter, wherein the preset second energy consumption parameter is the same as or different from the preset first energy consumption parameter.
  • step S600 can also have other specific implementation manners, and this application will be described in the following embodiments.
  • the model training method provided by the embodiment of the application can obtain the control parameters when the energy consumption is the lowest according to the energy consumption prediction model, and determine the model to obtain the control offset according to the current offset, and then according to the disturbance amount and the control offset Determine the actual offset; obtain a set of test control parameters for the air-conditioning system according to the actual offset and control parameters; obtain the energy consumption parameters of the air-conditioning system when operating under the set of test control parameters, and generate a test record ; Obtain training data according to the test records, and use the training data to train the current offset determination model.
  • the model training method of the present application can be trained to obtain the offset determination model, and the control offset of the model output is determined by the offset, which can be superimposed on the energy consumption prediction model to obtain the control parameters when the energy consumption is the lowest. , So as to obtain the target control parameters that can truly reduce the energy consumption of the air conditioning system.
  • this application uses the addition of disturbance before training to enrich the test data and improve the accuracy and applicability of the offset determination model.
  • the energy consumption parameter is the energy consumption ratio
  • the energy consumption ratio actual energy consumption of the air conditioning system/reference energy consumption of the air conditioning system.
  • S500 Obtain energy consumption parameters of the air-conditioning system when operating under the set of control parameters for the test, and generate a test record.
  • step S100 to step S500 have been specifically described in the embodiment shown in FIG. 1, and will not be repeated here.
  • This application may use explicit supervision or implicit supervision to train the current offset determination model using the training data.
  • step S600 may specifically include:
  • the current energy consumption ratio is determined to be the minimum energy consumption ratio output by the model The control offset
  • the actual offset that minimizes the energy consumption ratio output by the current energy consumption ratio determination model is used as the expected output of the current offset determination model, and the current offset determination model is trained, where
  • the input corresponding to the expected output is: a second environmental parameter group corresponding to the expected output in the test record.
  • control offset that minimizes the energy consumption ratio output by the current energy consumption ratio determination model is the training supervision value.
  • step S600 may specifically include:
  • the output of the current offset determination model is used as one of the inputs of the current energy consumption ratio determination model, and the current energy consumption ratio determination model outputs the minimum energy consumption ratio as the goal, obtained from test records Training data, using the training data to train the current offset determination model.
  • the implicit supervision method connects the offset determination model and the energy consumption ratio determination model in series.
  • the offset determination model can also be trained It becomes a machine model that can output a control offset that can effectively reduce energy consumption.
  • step S700 can be executed after step S500, and the execution sequence between step S700 and step S600 is not limited in this application.
  • an embodiment of the present application also provides a method for determining control parameters, which may include:
  • step S001 is the same as the processing performed in step S100 shown in FIG. 1, except that step S100 is performed in the model training phase, and step S001 is performed in the model application phase.
  • the model training phase and the model application phase can be alternately performed multiple times, that is, the offset determination model is trained multiple times in multiple appropriate time periods to achieve continuous optimization of the model.
  • the offset determination model after training can be used to obtain the control offset, thereby obtaining target control parameters that can effectively reduce energy consumption.
  • S003 Determine a target control parameter of the air conditioning system according to the obtained control parameter and the obtained control offset.
  • step S003 the method shown in FIG. 3 may further include:
  • the control parameter of the air conditioning system is set as the target control parameter.
  • the present application also provides a model training device, which may include: a first input unit 100, a second input unit 200, and a disturbance generating unit 300 , The test parameter obtaining unit 400, the test record generating unit 500 and the first training unit 600,
  • the first input unit 100 is configured to input the current first environmental parameter group of the air conditioning system into the energy consumption prediction model obtained by training, and obtain the control parameters when the energy consumption output by the energy consumption prediction model is the lowest, wherein:
  • the input of the energy consumption prediction model is the first environmental parameter group and the control parameter, and the output of the energy consumption prediction model is the energy consumption of the air conditioning system;
  • the first environmental parameter group may include: the required refrigeration capacity of the air-conditioning system and/or the uncontrolled temperature of the air-conditioning system, etc.
  • the control parameters may include at least one of the inlet water temperature of the air-conditioning system, the outlet water temperature of the air-conditioning system, the frequency of the water pump, and the flow rate of cold water.
  • energy consumption refers to the electrical energy consumed by the operation of the air-conditioning system, which can be determined according to the power of the air-conditioning system.
  • the second input unit 200 is configured to input the current second environmental parameter group into the current offset determination model to obtain the control offset output by the current offset determination model, wherein the offset
  • the input of the displacement determination model is the second environmental parameter group, and the output is the control offset
  • the second environmental parameter group may include outdoor environmental parameters and indoor environmental parameters, where the outdoor environmental parameters may include at least one of outdoor temperature, outdoor humidity, thermodynamic wet bulb temperature, wind direction, wind speed and other parameters.
  • indoor environmental parameters may include: indoor temperature, indoor maximum temperature, indoor humidity, and so on.
  • the indoor environment parameters may also include: the power of the indoor electronic devices, etc.
  • the disturbance amount generating unit 300 is configured to generate a disturbance amount, and determine an actual offset amount according to the disturbance amount and the control offset amount;
  • the test parameter obtaining unit 400 is configured to obtain a set of test control parameters of the air conditioning system according to the actual offset and the obtained control parameters;
  • the test record generating unit 500 is configured to obtain the energy consumption parameters of the air-conditioning system when the air-conditioning system is operating under the set of control parameters for the test, and generate a test record;
  • the first training unit 600 is configured to obtain training data according to test records, and use the training data to train the current offset determination model, where the training data includes: a second environmental parameter group and actual Offset.
  • each test record contains at least a second environmental parameter group, an actual offset, and energy consumption parameters
  • the first training unit 600 is specifically configured to:
  • the obtained actual offset is used as the expected output of the current offset determination model, and the current offset determination model is trained, wherein the input corresponding to the expected output is: The second environmental parameter group corresponding to the actual offset.
  • the energy consumption parameter is an energy consumption ratio
  • each test record contains at least a second environmental parameter group, a control parameter, an actual offset, and an energy consumption parameter.
  • the energy consumption ratio Actual energy consumption of the air-conditioning system/reference energy consumption of the air-conditioning system.
  • a second training unit for obtaining, by the test record generating unit 500, that the air-conditioning system is running under the test control parameters
  • the actual offset is used as the control offset
  • the current second environmental parameter group, the control parameter, the actual offset, and the energy consumption parameter are used as training Data
  • the current energy consumption ratio determination model is trained, wherein the input of the energy consumption ratio determination model is the second environmental parameter group, the control parameter and the control offset, and the output of the energy consumption ratio determination model is energy Consumption ratio.
  • the first training unit 600 is specifically configured to:
  • the current energy consumption ratio is determined to be the minimum energy consumption ratio output by the model The control offset
  • the actual offset that minimizes the energy consumption ratio output by the current energy consumption ratio determination model is used as the expected output of the current offset determination model, and the current offset determination model is trained, where
  • the input corresponding to the expected output is: a second environmental parameter group corresponding to the expected output in the test record.
  • the first training unit 600 is specifically configured to:
  • the output of the current offset determination model is used as one of the inputs of the current energy consumption ratio determination model, and the current energy consumption ratio determination model outputs the minimum energy consumption ratio as the goal, obtained from test records Training data, using the training data to train the current offset determination model.
  • the energy consumption parameter is an energy consumption ratio
  • the test record generating unit 500 obtains the energy consumption parameter of the air conditioning system when the air conditioning system is operating under the set of control parameters for the test, and the specific setting is:
  • energy consumption ratio actual energy consumption/reference energy consumption of the air conditioning system, the energy consumption ratio is calculated.
  • the model training device can obtain the control parameters at the lowest energy consumption according to the energy consumption prediction model, and determine the model to obtain the control offset according to the current offset, and then according to the disturbance amount and the control offset, Determine the actual offset; obtain a set of test control parameters for the air-conditioning system according to the actual offset and control parameters; obtain the energy consumption parameters of the air-conditioning system when operating under the set of test control parameters, and generate a test record; Test records to obtain training data, and use the training data to train the current offset determination model.
  • the model training method of the present application can be trained to obtain the offset determination model, and the control offset of the model output is determined by the offset, which can be superimposed on the energy consumption prediction model to obtain the control parameters when the energy consumption is the lowest. , So as to obtain the target control parameters that can truly reduce the energy consumption of the air conditioning system.
  • this application uses the addition of disturbance before training to enrich the test data and improve the accuracy and applicability of the offset determination model.
  • control parameter determination device may include: a control parameter obtaining unit 001, an offset obtaining unit 002, and The target parameter determination unit 003,
  • the control parameter obtaining unit 001 is configured to input the current first environmental parameter group of the air-conditioning system into the energy consumption prediction model obtained by training, and obtain the control parameters when the energy consumption output by the energy consumption prediction model is the lowest, wherein:
  • the input of the energy consumption prediction model is the first environmental parameter group and the control parameter, and the output of the energy consumption prediction model is the energy consumption of the air conditioning system;
  • the offset obtaining unit 002 is configured to input the current second environmental parameter group into the current offset determination model to obtain the control offset output by the current offset determination model, wherein the The input of the offset determination model is the second environmental parameter group, and the output is the control offset;
  • the target parameter determining unit 003 is configured to determine the target control parameter of the air conditioning system according to the obtained control parameter and the obtained control offset.
  • the model training device includes a processor and a memory.
  • the first input unit, the second input unit, the disturbance generating unit, the test parameter obtaining unit, the test record generating unit, and the first training unit are all stored in the memory as program units.
  • the processor executes the above-mentioned program unit stored in the memory to realize the corresponding function.
  • the control parameter determining device includes a processor and a memory.
  • the control parameter obtaining unit, the offset obtaining unit, and the target parameter determining unit are all stored in the memory as program units, and the program units stored in the memory are executed by the processor. To realize the corresponding function.
  • the processor contains the kernel, and the kernel calls the corresponding program unit from the memory.
  • One or more kernels can be set, and the kernel parameters are adjusted to train the model and/or determine the control parameters.
  • the embodiment of the present application provides a storage medium on which a program is stored, and when the program is executed by a processor, the model training method and/or the control parameter determination method are implemented.
  • An embodiment of the present application provides a processor configured to run a program, wherein the model training method and/or the control parameter determination method are executed when the program is running.
  • an embodiment of the present application provides a device 70.
  • the device includes at least one processor 701, and at least one memory 702 and a bus 703 connected to the processor 701; wherein the processor 701 and the memory 702 pass through the bus 703 completes mutual communication; the processor 701 is configured to call program instructions in the memory 702 to execute the above-mentioned model training method and/or control parameter determination method.
  • the device 70 herein may be a server, PC, PAD, mobile phone, etc.
  • the application also provides a computer program product, which when executed on a data processing device, is suitable for executing a program initialized with the steps of the above-mentioned model training method and/or control parameter determination method.
  • the device includes one or more processors (CPUs), memory, and buses.
  • the device may also include input/output interfaces, network interfaces, and so on.
  • the memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one Memory chip.
  • RAM random access memory
  • ROM read-only memory
  • flash RAM flash memory
  • the memory is an example of a computer-readable medium.
  • Computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology.
  • the information can be computer-readable instructions, data structures, program modules, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, Magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. According to the definition in this article, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.
  • this application can be provided as methods, systems, or computer program products. Therefore, this application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Software Systems (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Signal Processing (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Economics (AREA)
  • Combustion & Propulsion (AREA)
  • Mechanical Engineering (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Air Conditioning Control Device (AREA)

Abstract

一种模型训练方法、控制参数确定方法及装置，可以根据能耗预测模型得到能耗最低时的控制参数，并根据当前的偏移量确定模型获得控制偏移量，进而根据扰动量与控制偏移量，确定实际偏移量；根据实际偏移量与获得的控制参数，获得空调系统的一组测试用控制参数（S400）；获得空调系统在该组测试用控制参数下运行时的能耗参数，生成一条测试记录（S500）；根据测试记录获得训练数据，使用训练数据对当前的偏移量确定模型进行训练（S600）。通过本方法可以训练得到偏移量确定模型，通过该偏移量确定模型输出的控制偏移量，本方法可以得到能够真正降低空调系统的能耗的目标控制参数。本方法通过扰动量提高了偏移量确定模型的准确性和适用性。

Description

模型训练方法、控制参数确定方法及装置
本申请要求于2019年10月31日提交中国专利局、申请号为201911055197.8、发明名称为“模型训练方法、控制参数确定方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及节能降耗技术领域，尤其涉及模型训练方法、控制参数确定方法及装置。
背景技术
空调系统的节能降耗是节能降耗领域的重要组成部分。
现有的对空调系统的降耗技术采用了能耗预测模型来获得能耗预测模型输出的能耗最低时的控制参数。
虽然对于能耗预测模型而言，现有技术获得的控制参数是使得能耗预测模型输出的功耗最低时的控制参数。但是在实际应用中，由于多种原因，现有技术通过能耗预测模型获得的控制参数在施加到空调系统上时，却往往无法达到有效降低空调系统能耗的效果。
发明内容
鉴于上述问题，本申请提供一种克服上述问题或者至少部分地解决上述问题的模型训练方法、控制参数确定方法及装置，方案如下：
一种模型训练方法，包括：
将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
生成扰动量，根据所述扰动量与所述控制偏移量，确定实际偏移量；
根据所述实际偏移量与获得的所述控制参数，获得所述空调系统的一组测试用控制参数；
获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，生成一条测试记录；
根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，其中，所述训练数据包括：第二环境参数组和实际偏移量。
可选的，每条所述测试记录中均至少对应保存有：第二环境参数组、实际偏移量和能耗参数，所述根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练包括：
从测试记录中获得与满足预设的能耗要求的能耗参数相对应的第二环境参数组和实际偏移量；
将获得的实际偏移量作为所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训练，其中，所述期望输出对应的输入为：与所述获得的实际偏移量对应的第二环境参数组。
可选的，所述能耗参数为能耗比，每条所述测试记录中均至少对应保存有：第二环境参数组、控制参数、实际偏移量和能耗参数，所述能耗比＝空调系统实际能耗/空调系统基准能耗，在所述获得所述空调系统在所述测试用控制参数下运行时的能耗参数后，所述方法还包括：
将所述实际偏移量作为控制偏移量，将所述当前的第二环境参数组、所述控制参数、所述实际偏移量和所述能耗参数作为训练数据，对当前的能耗比确定模型进行训练，其中，所述能耗比确定模型的输入为第二环境参数组、控制参数和控制偏移量，所述能耗比确定模型的输出为能耗比。
可选的，所述根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，包括：
将所述实际偏移量作为控制偏移量，获得输入为所述测试记录中的第二环境参数组和实际偏移量时、使得所述当前的能耗比确定模型输出的能耗比最小的控制偏移量；
将使得所述当前的能耗比确定模型输出的能耗比最小的实际偏移量作为 所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训练，其中，所述期望输出对应的输入为：在所述测试记录中与所述期望输出对应的第二环境参数组。
可选的，所述根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，包括：
将所述当前的偏移量确定模型的输出作为所述当前的能耗比确定模型的输入之一，以所述当前的能耗比确定模型输出的能耗比最小为目标，从测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练。
可选的，所述能耗参数为能耗比，所述获得所述空调系统在所述测试用控制参数下运行时的能耗参数，包括：
将所述当前的第二环境参数组输入训练得到的环境基准能耗预测模型中，获得所述训练得到的环境基准能耗预测模型输出的空调系统基准能耗；
获得所述空调系统在所述测试用控制参数下运行时的实际能耗；
根据公式：能耗比＝实际能耗/空调系统基准能耗，计算得到所述能耗比。
一种控制参数确定方法，包括：
将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
根据获得的所述控制参数与获得的所述控制偏移量，确定所述空调系统的目标控制参数。
一种模型训练装置，包括：第一输入单元、第二输入单元、扰动量生成单元、测试参数获得单元、测试记录生成单元和第一训练单元，
所述第一输入单元，用于将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数， 其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
所述第二输入单元，用于将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
所述扰动量生成单元，用于生成扰动量，根据所述扰动量与所述控制偏移量，确定实际偏移量；
所述测试参数获得单元，用于根据所述实际偏移量与获得的所述控制参数，获得所述空调系统的一组测试用控制参数；
所述测试记录生成单元，用于获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，生成一条测试记录；
所述第一训练单元，用于根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，其中，所述训练数据包括：第二环境参数组和实际偏移量。
一种控制参数确定装置，包括：控制参数获得单元、偏移量获得单元和目标参数确定单元，
所述控制参数获得单元，用于将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
所述偏移量获得单元，用于将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
所述目标参数确定单元，用于根据获得的所述控制参数与获得的所述控制偏移量，确定所述空调系统的目标控制参数。
一种存储介质，所述存储介质包括存储的程序，其中，在所述程序运行时控制所述存储介质所在设备执行如上述的任一种模型训练方法和/或上述的控制参数确定方法。
一种设备，所述设备包括至少一个处理器、以及与所述处理器连接的至少一个存储器、总线；其中，所述处理器、所述存储器通过所述总线完成相互间的通信；所述处理器用于调用所述存储器中的程序指令，以执行如上述的任一种模型训练方法和/或上述的控制参数确定方法。
借由上述技术方案，本申请提供的一种模型训练方法、控制参数确定方法及装置，可以根据能耗预测模型得到能耗最低时的控制参数，并根据当前的偏移量确定模型获得控制偏移量，进而根据扰动量与控制偏移量，确定实际偏移量；根据实际偏移量与控制参数，获得空调系统的一组测试用控制参数；获得空调系统在该组测试用控制参数下运行时的能耗参数，生成一条测试记录；根据测试记录获得训练数据，使用训练数据对当前的偏移量确定模型进行训练。通过本申请的模型训练方法可以训练得到偏移量确定模型，通过该偏移量确定模型输出的控制偏移量，本申请可以将其叠加到能耗预测模型得到能耗最低时的控制参数上，从而得到能够真正降低空调系统的能耗的目标控制参数。同时，本申请在训练前，通过扰动量的添加，使得测试数据更为丰富，提高了偏移量确定模型的准确性和适用性。
上述说明仅是本申请技术方案的概述，为了能够更清楚了解本申请的技术手段，而可依照说明书的内容予以实施，并且为了让本申请的上述和其它目的、特征和优点能够更明显易懂，以下特举本申请的具体实施方式。
附图说明
通过阅读下文优选实施方式的详细描述，各种其他的优点和益处对于本领域普通技术人员将变得清楚明了。附图仅用于示出优选实施方式的目的，而并不认为是对本申请的限制。而且在整个附图中，用相同的参考符号表示相同的部件。在附图中：
图1示出了本申请实施例提供的一种模型训练方法的流程图；
图2示出了本申请实施例提供的另一种模型训练方法的流程图；
图3示出了本申请实施例提供的一种控制参数确定方法的流程图；
图4示出了本申请实施例提供的一种模型训练装置的结构示意图；
图5示出了本申请实施例提供的一种控制参数确定装置的结构示意图；
图6示出了本申请实施例提供的一种设备的结构示意图。
具体实施方式
下面将参照附图更详细地描述本公开的示例性实施例。虽然附图中显示了本公开的示例性实施例，然而应当理解，可以以各种形式实现本公开而不应被这里阐述的实施例所限制。相反，提供这些实施例是为了能够更透彻地理解本公开，并且能够将本公开的范围完整的传达给本领域的技术人员。
能耗预测模型是通过将采集的历史数据作为训练数据进行训练后得到的机器模型，其中，历史数据包括：相对应的第一环境参数组、控制参数和空调系统的能耗。其中，第一坏境参数组可以包括：空调系统的需求制冷量和/或空调系统非控温度等。空调系统的需求制冷量可以根据室内外的温度和湿度等环境参数确定。空调系统非控温度可以包括：蒸发温度、冷凝温度空调系统运行过程中未控制的温度等。由于历史数据的覆盖程度等原因，历史数据与当前时刻的相应数据往往有较大的差异，这就使得通过能耗预测模型得到的控制参数往往并不是能有效降低空调系统能耗的控制参数，也往往不是能使得空调系统能耗最低的最优控制参数。当然，除由于历史数据的原因外，能耗预测模型本身的结构设计等原因也会导致通过能耗预测模型得到的控制参数往往并不是能有效降低空调系统能耗的控制参数。
对于上述问题，现有的技术人员普遍都是通过完善能耗预测模型的方式来努力提高能耗预测模型的准确性。例如：采集更多的历史数据，优化能耗预测模型的结构等。与现有的技术人员不同，本申请的发明人的研究方向放在了根据能耗预测模型获得的能耗最低时的控制参数与空调系统实际能耗最低时的最优控制参数的控制偏移量上。基于此，本申请的发明人提出了偏移量确定模型以确定该控制偏移量。该偏移量确定模型的输入为第二环境参数组，输出为控制偏移量。在根据能耗预测模型获得的能耗最低时的控制参数后，只需要再根据本申请的偏移量确定模型确定控制偏移量，就可以根据上述控制参数和控制偏移量获得能够真正降低空调系统的能耗的目标控制参数。
基于此，本申请首先提供了一种模型训练方法，以对偏移量确定模型进行训练。通过该模型训练方法，本申请可以获得偏移量确定模型，还可以不断对 偏移量确定模型进行改进。
如图1所示，本申请实施例提供了一种模型训练方法，可以包括：
S100、将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为第一环境参数组和控制参数，所述能耗预测模型的输出为所述空调系统的能耗。
可选的，由于空调系统的需求制冷量可以根据室内外的温度和湿度等环境参数确定，因此除包括空调系统非控温度外，第一环境参数组还可以包括：室外温度、室外湿度、热力学湿球温度、风向、风速、室内温度、室内最大温度、室内湿度等环境参数中的至少一种。
其中，控制参数可以包括：空调系统的进水温度、空调系统的出水温度、水泵的频率、冷水流量等参数中的至少一种。
其中，能耗是指空调系统运行所消耗的电能，可以根据空调系统的功率确定。
S200、将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为第二环境参数组，输出为所述控制偏移量。
其中，第二环境参数组可以包括：室外环境参数和室内环境参数，其中，室外环境参数可以包括：室外温度、室外湿度、热力学湿球温度、风向、风速等参数中的至少一种。其中，室内环境参数可以包括：室内温度、室内最大温度、室内湿度等。可选的，当室内设置有较多的电子设备时(如机房内有大量电子设备)，室内环境参数还可以包括：室内电子设备的功率等。
其中，第二环境参数组和第一环境参数组可以具有交集，也可以不具有交集。可选的，第二环境参数组可以为第一环境参数组的真子集。
其中，步骤S100和步骤S200之间的执行先后关系本申请不做限定。
可以理解的是，由于第二环境参数组、控制参数可以分别包括多种参数，因此本申请可以使用向量来表示第二环境参数组和控制参数。例如：第二环境参数组为[35，90，25，30，70]，该一维向量代表：室外温度35度，室外湿 度90％，室内温度25度，室内最大温度30度，室内湿度70％。当然，还可以用多维向量表示第二环境参数组和控制参数，本申请在此不做限定。再如：控制参数为[12，17，40，558]，该一维向量代表：进水温度12度，出水温度17，水泵的频率为40Hz，冷水流量为558立方米/每小时。
具体的，在当前的偏移量确定模型进行第一次训练前，当前的偏移量确定模型为初始偏移量确定模型。其中，初始偏移量确定模型输出的控制偏移量均为0。
当初始偏移量确定模型进行了训练后，该模型的输出的控制偏移量一般不会再全部为0。
其中，偏移量确定模型输出的控制偏移量的向量的结构与根据能耗预测模型得到的控制参数的向量的结构可以相同，例如：可以均为4×4的结构。当然，偏移量确定模型输出的控制偏移量的向量中各元素的含义与根据能耗预测模型得到的控制参数的向量中相同位置的元素的含义也相同。例如，对于上述控制参数的向量[12，17，40，558]，步骤S200获得的控制偏移量可以为[+0.5，-0.3，+4，-8]，该一维向量代表：进水温度的偏移量为+0.5度，出水温度的偏移量为-0.3度，水泵的频率的偏移量为+4Hz，冷水流量的偏移量为-8立方米/每小时。
需要说明的是，控制偏移量除与第二环境参数组有关外，还与根据能耗预测模型得到的控制参数有关。但是，由于第一环境参数组中的空调系统的需求制冷量也是根据第二环境参数组得到的，因此能耗预测模型的间接根据第二环境参数组得到了控制参数。为了避免过耦合，本申请中的偏移量确定模型的输入只有第二环境参数组。
S300、生成扰动量，根据所述扰动量与所述控制偏移量，确定实际偏移量。
其中，本申请可以通过多种方式生成扰动量，例如通过随机数生成算法生成一个随机数，将该随机数作为扰动量。再如：通过正交测试表生成扰动量。具体的，步骤S300确定的实际偏移量只是一个假设的实际偏移量，并不代表真正的实际偏移量。
其中，扰动量也可以为向量，偏移量确定模型输出的控制偏移量的向量的 结构与扰动量的向量的结构可以相同，当然，偏移量确定模型输出的控制偏移量的向量中各元素的含义与扰动量的向量中相同位置的元素的含义也相同。例如：当步骤S200获得的控制偏移量为上述的[+0.5，-0.3，+4，-8]时，步骤S300生成的扰动量可以为：[+0.1，-0.1，+0.4，-2]，该一维向量代表：进水温度的扰动量为+0.1度，出水温度的扰动量为-0.1度，水泵的频率的扰动量为+0.4Hz，冷水流量的扰动量为-2立方米/每小时。当然，扰动量还可以用比例来表示，如步骤S300生成的扰动量可以为：[+5％，-5％，+8％，-2％]，该一维向量代表：进水温度的扰动量为+5％，出水温度的扰动量为-5％，水泵的频率的扰动量为+8％，冷水流量的扰动量为-2％。
当扰动量中的元素代表控制量的数值时，可以将扰动量与所述控制偏移量相加，得到实际偏移量。例如：将[+0.5，-0.3，+4，-8]与[+0.1，-0.1，+0.4，-2]相加，获得实际偏移量为[+0.6，-0.4，+4.4，-10]。
当扰动量中的元素代表控制量的比例时，可以根据比例乘法在控制偏移量上增加或减少相应的比例。例如：在[+0.5，-0.3，+4，-8]基础上，按照[+5％，-5％，+8％，-2％]进行浮动，获得[+0.525，-0.285，+4.32，-7.84]。
可选的，扰动量可以在预设的数值范围或比例范围内，例如：进水温度的扰动量在±1度的范围内，冷水流量的扰动量在±50立方米/每小时的范围内。
由于每次生成的扰动量都有可能是不同的，因此通过在控制偏移量上增加扰动量，本申请可以获得更多的测试数据，且为测试数据添加了随机性。同时，由于扰动量在预设的数值范围内，因此不会使得空调系统运行在明显不合理的情况下，不会给空调系统带来太大的运行负担。同时，较小的扰动量也提供了较小的测试粒度，使得测试数据的覆盖度变高，进而使得训练得到的偏移量确定模型的准确性更高。
进一步，为了防止产生的扰动量相同而进行重复测试，本申请可以将之前产生的扰动量中与当前的第二环境参数组对应的扰动量确定为待比较扰动量，将待比较扰动量分别与本次产生的扰动量进行对比，从而确定是否已经在当前的第二环境参数组下进行过本次产生的扰动量的测试。如果有至少一个待比较扰动量与本次产生的扰动量相等，则丢弃本次产生的扰动量，再次产生一个新 的扰动量并继续进行上述对比。如果各待比较扰动量均与本次产生的扰动量不相等，则可以根据本次产生的扰动量与所述控制偏移量，确定实际偏移量并执行后续步骤。
S400、根据所述实际偏移量与获得的所述控制参数，获得所述空调系统的一组测试用控制参数。
可选的，本申请可以将实际偏移量与获得的所述控制参数相加，获得空调系统的一组测试用控制参数。例如：当实际偏移量为[+0.6，-0.4，+4.4，-10]且控制参数为[12，17，40，558]时，将二者叠加得到一组测试用控制参数为[12.6，16.6，44.4，548]。
S500、获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，生成一条测试记录。
其中，每条所述测试记录中均对应保存有：第二环境参数组、控制参数、控制偏移量、扰动量和能耗参数中的至少一个。
在获得一组测试用控制参数后，本申请可以控制空调系统的控制参数为该组测试用控制参数，例如：当测试用控制参数为[12.6，16.6，44.4，548]时，控制空调系统的进水温度为12.6度，出水温度为16.6度，水泵的频率为44.4Hz，冷水流量为548立方米/每小时。
在实际应用中，在控制空调系统的控制参数为该组测试用控制参数后，可以等待一段时间后，再对能耗参数进行采集。这是由于控制参数变动后，空调系统需要经过一段时间后才会进入一个相对稳定的状态，该相对稳定的状态的功耗参数的实际意义更大。
在实际应用中，本申请可以使用一个表格来记录各条测试记录，该表格可以如表1所示：
表1
Figure PCTCN2020100520-appb-000001
Figure PCTCN2020100520-appb-000002
其中，表1中能耗参数为能耗比，所述能耗比＝空调系统实际能耗/空调系统基准能耗。本申请在对空调系统进行测试时，可以通过实际测量的方式来确定空调系统实际能耗。本申请可以通过将所述当前的第二环境参数组输入训练得到的环境基准能耗预测模型中，获得所述训练得到的环境基准能耗预测模型输出的空调系统基准能耗。其中，环境基准能耗预测模型是通过训练数据(该训练数据包括相对应的第二环境参数组和空调系统基准能耗)训练得到的机器模型，其输入为第二环境参数组，输出为空调系统基准能耗。
可选的，在其他实施例中，步骤S500可以具体包括：
将所述当前的第二环境参数组输入训练得到的环境基准能耗预测模型中，获得所述训练得到的环境基准能耗预测模型输出的空调系统基准能耗；
获得所述空调系统在所述测试用控制参数下运行时的实际能耗；
根据公式：能耗比＝实际能耗/空调系统基准能耗，计算得到所述能耗比。
在实际应用中，对于某个第二环境参数组下，本申请可以生成多个不同的扰动量，并将这些扰动量分别添加当前的偏移量确定模型根据该第二环境参数组得到的控制偏移量上，到进而得到多个不同的实际偏移量。然后，本申请将这些实际偏移量分别添加到能耗预测模型根据该第二环境参数组输出的能耗最低时的控制参数上，从而获得多组测试用控制参数。这些多组测试用控制参数都是在同一个第二环境参数组下的测试用控制参数。
本申请可以控制空调系统在不同的时间段分别按照上述多组测试用控制参数运行，从而可以采集到多条测试记录。这些测试记录中的：第二环境参数组、控制参数和控制偏移量均相同，但是扰动量不同。这样，本申请就可以对同一个环境下多个不同扰动所带来的能耗参数进行了测试，产生了较高的数据覆盖效果。
可以理解的是，本申请可以在多种情况下根据测试记录来对当前的偏移量确定模型进行训练。上述多种情况可以包括：
情况一、测试记录的条数少于第一预设条数。
此情况下，测试记录还比较少，需要继续进行测试以积累数据。
当然，情况一可以具体为：预设的测试时间段内的测试记录的条数少于第一预设条数。
情况二、当前的第二环境参数组超出已获得的测试记录中的第二环境参数组的覆盖范围。
例如：已获得的测试记录中的第二环境参数组中的最低的室外温度为零下5度，当前的第二环境参数组中的室外温度为零下7度，则当前的第二环境参数组超出已获得的测试记录中的第二环境参数组的覆盖范围。
通过情况二，本申请可以有效采集更多环境下的测试数据，以提高测试数据的覆盖率，提高机器模型的准确度和适用性。
情况三、已获得的测试记录中的与当前的第二环境参数组对应的能耗参数高于预设的第一能耗参数。
其中，预设的第一能耗参数可以为：预设能耗比，当某能耗比高于预设能耗比时，说明该能耗比较高，无法实现较好的降低能耗的作用。需要继续测试。
情况四、已获得的测试记录中的与当前的第二环境参数组对应的能耗参数高于预设的第一能耗参数，且已获得的当前的第二环境参数组对应的测试记录的条数少于第二预设条数，其中，第二预设条数大于第一预设条数。
通过上述第二预设条数的限制，可以防止进行多过的测试。因为过多的测试会影响空调系统的正常使用。
情况五、已获得的测试记录中的与当前的第二环境参数组对应的能耗参数高于预设的第一能耗参数，且测试记录的条数少于第二预设条数，其中，第二预设条数大于第一预设条数。
通过上述第二预设条数的限制，可以防止进行多过的测试。因为过多的测试会影响空调系统的正常使用。
S600、根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，其中，所述训练数据包括：第二环境参数组和实际偏移量。
其中，每条所述测试记录中均至少对应保存有：第二环境参数组、实际偏移量和能耗参数，步骤S600可以具体包括：
从测试记录中获得与满足预设的能耗要求的能耗参数相对应的第二环境参数组和实际偏移量；
将获得的实际偏移量作为所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训练，其中，所述期望输出对应的输入为：与所述获得的实际偏移量对应的第二环境参数组。
可以理解的是，通过添加扰动量后的测试，本申请可以获得多条测试记录，但是并不是所有的测试记录中的能耗参数都很好。为了尽可能的获得最优偏移量，本申请可以对测试记录进行筛选，从中筛选出满足预设的能耗要求的测试记录，然后将筛选出的测试记录中的第二环境参数组和实际偏移量作为训练数据对所述当前的偏移量确定模型进行训练。
上述预设的能耗要求可以为能耗参数不高于预设的第二能耗参数，其中，预设的第二能耗参数与预设的第一能耗参数相同或不同。
当然，步骤S600也可以有其他具体实施方式，本申请将在下述实施例中说明。
本申请实施例提供的一种模型训练方法，可以根据能耗预测模型得到能耗最低时的控制参数，并根据当前的偏移量确定模型获得控制偏移量，进而根据扰动量与控制偏移量，确定实际偏移量；根据实际偏移量与控制参数，获得空调系统的一组测试用控制参数；获得空调系统在该组测试用控制参数下运行时的能耗参数，生成一条测试记录；根据测试记录获得训练数据，使用训练数据对当前的偏移量确定模型进行训练。通过本申请的模型训练方法可以训练得到偏移量确定模型，通过该偏移量确定模型输出的控制偏移量，本申请可以将其叠加到能耗预测模型得到能耗最低时的控制参数上，从而得到能够真正降低空调系统的能耗的目标控制参数。同时，本申请在训练前，通过扰动量的添加，使得测试数据更为丰富，提高了偏移量确定模型的准确性和适用性。
如图2所示，本申请实施例提供的另一种模型训练方法中，能耗参数为能耗比，所述能耗比＝空调系统实际能耗/空调系统基准能耗，该方法可以包括：
S100、将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
S200、将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
S300、生成扰动量，根据所述扰动量与所述控制偏移量，确定实际偏移量；
S400、根据所述实际偏移量与获得的所述控制参数，获得所述空调系统的一组测试用控制参数；
S500、获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，生成一条测试记录。
其中，步骤S100至步骤S500已在图1所示实施例中具体说明，在此不再赘述。
S600、根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，其中，所述训练数据包括：第二环境参数组和实际偏移量。
本申请可以使用显式监督或隐式监督的方式来使用所述训练数据对所述当前的偏移量确定模型进行训练。
当使用显式监督的方式时，可选的，步骤S600可以具体包括：
将所述实际偏移量作为控制偏移量，获得输入为所述测试记录中的第二环境参数组和实际偏移量时、使得所述当前的能耗比确定模型输出的能耗比最小的控制偏移量；
将使得所述当前的能耗比确定模型输出的能耗比最小的实际偏移量作为所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训练，其中，所述期望输出对应的输入为：在所述测试记录中与所述期望输出对应的第二环境参数组。
其中，使得所述当前的能耗比确定模型输出的能耗比最小的控制偏移量即 为训练监督值。
当使用隐式监督的方式时，可选的，步骤S600可以具体包括：
将所述当前的偏移量确定模型的输出作为所述当前的能耗比确定模型的输入之一，以所述当前的能耗比确定模型输出的能耗比最小为目标，从测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练。
隐式监督的方式将偏移量确定模型和能耗比确定模型进行了串联，通过以所述当前的能耗比确定模型输出的能耗比最小为目标，同样可以将偏移量确定模型训练成为可以输出能够有效降低能耗的控制偏移量的机器模型。
S700、将所述实际偏移量作为控制偏移量，将所述当前的第二环境参数组、所述控制参数、所述实际偏移量和所述能耗参数作为训练数据，对当前的能耗比确定模型进行训练，其中，所述能耗比确定模型的输入为第二环境参数组、控制参数和控制偏移量，所述能耗比确定模型的输出为能耗比。
其中，步骤S700在步骤S500之后执行即可，步骤S700与步骤S600之间的执行顺序本申请不做限定。
如图3所示，本申请实施例还提供了一种控制参数确定方法，可以包括：
S001、将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
其中，步骤S001与图1所示步骤S100执行的处理相同，区别在于步骤S100在模型训练阶段中执行，而步骤S001在模型应用阶段中执行。
当然，模型训练阶段和模型应用阶段可以多次交替进行，即：在合适的多个时间段来多次对偏移量确定模型进行训练，以实现模型的持续优化。当不需要进行训练时，可以使用训练后的偏移量确定模型来获得控制偏移量，进而获得能够有效降低能耗的目标控制参数。
S002、将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为第二环境参数组，输出为所述控制偏移量；
S003、根据获得的所述控制参数与获得的所述控制偏移量，确定所述空调系统的目标控制参数。
进一步，步骤S003后，图3所示方法还可以包括：
将空调系统的控制参数设置为所述目标控制参数。
与本申请实施例提供的模型训练方法相对应，如图4所示，本申请还提供了一种模型训练装置，可以包括：第一输入单元100、第二输入单元200、扰动量生成单元300、测试参数获得单元400、测试记录生成单元500和第一训练单元600，
所述第一输入单元100，用于将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
其中，第一坏境参数组可以包括：空调系统的需求制冷量和/或空调系统非控温度等。
其中，控制参数可以包括：空调系统的进水温度、空调系统的出水温度、水泵的频率、冷水流量等参数中的至少一种。
其中，能耗是指空调系统运行所消耗的电能，可以根据空调系统的功率确定。
所述第二输入单元200，用于将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
其中，第二环境参数组可以包括：室外环境参数和室内环境参数，其中，室外环境参数可以包括：室外温度、室外湿度、热力学湿球温度、风向、风速等参数中的至少一种。其中，室内环境参数可以包括：室内温度、室内最大温度、室内湿度等。可选的，当室内设置有较多的电子设备时(如机房内有大量电子设备)，室内环境参数还可以包括：室内电子设备的功率等。
所述扰动量生成单元300，用于生成扰动量，根据所述扰动量与所述控制偏移量，确定实际偏移量；
所述测试参数获得单元400，用于根据所述实际偏移量与获得的所述控制参数，获得所述空调系统的一组测试用控制参数；
所述测试记录生成单元500，用于获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，生成一条测试记录；
所述第一训练单元600，用于根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，其中，所述训练数据包括：第二环境参数组和实际偏移量。
可选的，每条所述测试记录中均至少对应保存有：第二环境参数组、实际偏移量和能耗参数，所述第一训练单元600具体用于：
从测试记录中获得与满足预设的能耗要求的能耗参数相对应的第二环境参数组和实际偏移量；
将获得的实际偏移量作为所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训练，其中，所述期望输出对应的输入为：与所述获得的实际偏移量对应的第二环境参数组。
可选的，所述能耗参数为能耗比，每条所述测试记录中均至少对应保存有：第二环境参数组、控制参数、实际偏移量和能耗参数，所述能耗比＝空调系统实际能耗/空调系统基准能耗，图4所示装置还包括：第二训练单元，用于在所述测试记录生成单元500获得所述空调系统在所述测试用控制参数下运行时的能耗参数后，将所述实际偏移量作为控制偏移量，将所述当前的第二环境参数组、所述控制参数、所述实际偏移量和所述能耗参数作为训练数据，对当前的能耗比确定模型进行训练，其中，所述能耗比确定模型的输入为第二环境参数组、控制参数和控制偏移量，所述能耗比确定模型的输出为能耗比。
可选的，所述第一训练单元600具体用于：
将所述实际偏移量作为控制偏移量，获得输入为所述测试记录中的第二环境参数组和实际偏移量时、使得所述当前的能耗比确定模型输出的能耗比最小的控制偏移量；
将使得所述当前的能耗比确定模型输出的能耗比最小的实际偏移量作为所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训 练，其中，所述期望输出对应的输入为：在所述测试记录中与所述期望输出对应的第二环境参数组。
可选的，所述第一训练单元600具体用于：
将所述当前的偏移量确定模型的输出作为所述当前的能耗比确定模型的输入之一，以所述当前的能耗比确定模型输出的能耗比最小为目标，从测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练。
可选的，所述能耗参数为能耗比，所述测试记录生成单元500获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，具体设置为：
将所述当前的第二环境参数组输入训练得到的环境基准能耗预测模型中，获得所述训练得到的环境基准能耗预测模型输出的空调系统基准能耗；
获得所述空调系统在所述测试用控制参数下运行时的实际能耗；
根据公式：能耗比＝实际能耗/空调系统基准能耗，计算得到所述能耗比。
本申请提供的一种模型训练装置，可以根据能耗预测模型得到能耗最低时的控制参数，并根据当前的偏移量确定模型获得控制偏移量，进而根据扰动量与控制偏移量，确定实际偏移量；根据实际偏移量与控制参数，获得空调系统的一组测试用控制参数；获得空调系统在该组测试用控制参数下运行时的能耗参数，生成一条测试记录；根据测试记录获得训练数据，使用训练数据对当前的偏移量确定模型进行训练。通过本申请的模型训练方法可以训练得到偏移量确定模型，通过该偏移量确定模型输出的控制偏移量，本申请可以将其叠加到能耗预测模型得到能耗最低时的控制参数上，从而得到能够真正降低空调系统的能耗的目标控制参数。同时，本申请在训练前，通过扰动量的添加，使得测试数据更为丰富，提高了偏移量确定模型的准确性和适用性。
与图3所示方法实施例相对应，本申请还提供了一种控制参数确定装置，如图5所示，该控制参数确定装置可以包括：控制参数获得单元001、偏移量获得单元002和目标参数确定单元003，
所述控制参数获得单元001，用于将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参 数，所述能耗预测模型的输出为所述空调系统的能耗；
所述偏移量获得单元002，用于将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
所述目标参数确定单元003，用于根据获得的所述控制参数与获得的所述控制偏移量，确定所述空调系统的目标控制参数。
所述模型训练装置包括处理器和存储器，上述第一输入单元、第二输入单元、扰动量生成单元、测试参数获得单元、测试记录生成单元和第一训练单元等均作为程序单元存储在存储器中，由处理器执行存储在存储器中的上述程序单元来实现相应的功能。
所述控制参数确定装置包括处理器和存储器，上述控制参数获得单元、偏移量获得单元和目标参数确定单元等均作为程序单元存储在存储器中，由处理器执行存储在存储器中的上述程序单元来实现相应的功能。
处理器中包含内核，由内核去存储器中调取相应的程序单元。内核可以设置一个或以上，通过调整内核参数来训练模型和/或确定控制参数。
本申请实施例提供了一种存储介质，其上存储有程序，该程序被处理器执行时实现所述模型训练方法和/或控制参数确定方法。
本申请实施例提供了一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行所述模型训练方法和/或控制参数确定方法。
如图6所示，本申请实施例提供了一种设备70，设备包括至少一个处理器701、以及与处理器701连接的至少一个存储器702、总线703；其中，处理器701、存储器702通过总线703完成相互间的通信；处理器701用于调用存储器702中的程序指令，以执行上述的模型训练方法和/或控制参数确定方法。本文中的设备70可以是服务器、PC、PAD、手机等。
本申请还提供了一种计算机程序产品，当在数据处理设备上执行时，适于执行初始化有上述的模型训练方法和/或控制参数确定方法步骤的程序。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和 /或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
在一个典型的配置中，设备包括一个或多个处理器(CPU)、存储器和总线。设备还可以包括输入/输出接口、网络接口等。
存储器可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)，存储器包括至少一个存储芯片。存储器是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
还需要说明的是，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白，本申请的实施例可提供为方法、系统或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例或结合软件和 硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
以上仅为本申请的实施例而已，并不用于限制本申请。对于本领域技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本申请的权利要求范围之内。

Claims (11)

  1. 一种模型训练方法，其特征在于，包括：
    将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
    将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
    生成扰动量，根据所述扰动量与所述控制偏移量，确定实际偏移量；
    根据所述实际偏移量与获得的所述控制参数，获得所述空调系统的一组测试用控制参数；
    获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，生成一条测试记录；
    根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，其中，所述训练数据包括：第二环境参数组和实际偏移量。
  2. 根据权利要求1所述的方法，其特征在于，每条所述测试记录中均至少对应保存有：第二环境参数组、实际偏移量和能耗参数，所述根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练包括：
    从测试记录中获得与满足预设的能耗要求的能耗参数相对应的第二环境参数组和实际偏移量；
    将获得的实际偏移量作为所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训练，其中，所述期望输出对应的输入为：与所述获得的实际偏移量对应的第二环境参数组。
  3. 根据权利要求1所述的方法，其特征在于，所述能耗参数为能耗比，每条所述测试记录中均至少对应保存有：第二环境参数组、控制参数、实际偏移量和能耗参数，所述能耗比＝空调系统实际能耗/空调系统基准能耗，在所述 获得所述空调系统在所述测试用控制参数下运行时的能耗参数后，所述方法还包括：
    将所述实际偏移量作为控制偏移量，将所述当前的第二环境参数组、所述控制参数、所述实际偏移量和所述能耗参数作为训练数据，对当前的能耗比确定模型进行训练，其中，所述能耗比确定模型的输入为第二环境参数组、控制参数和控制偏移量，所述能耗比确定模型的输出为能耗比。
  4. 根据权利要求3所述的方法，其特征在于，所述根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，包括：
    将所述实际偏移量作为控制偏移量，获得输入为所述测试记录中的第二环境参数组和实际偏移量时、使得所述当前的能耗比确定模型输出的能耗比最小的控制偏移量；
    将使得所述当前的能耗比确定模型输出的能耗比最小的实际偏移量作为所述当前的偏移量确定模型的期望输出，对所述当前的偏移量确定模型进行训练，其中，所述期望输出对应的输入为：在所述测试记录中与所述期望输出对应的第二环境参数组。
  5. 根据权利要求3所述的方法，其特征在于，所述根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，包括：
    将所述当前的偏移量确定模型的输出作为所述当前的能耗比确定模型的输入之一，以所述当前的能耗比确定模型输出的能耗比最小为目标，从测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练。
  6. 根据权利要求1至5中任一项所述的方法，其特征在于，所述能耗参数为能耗比，所述获得所述空调系统在所述测试用控制参数下运行时的能耗参数，包括：
    将所述当前的第二环境参数组输入训练得到的环境基准能耗预测模型中，获得所述训练得到的环境基准能耗预测模型输出的空调系统基准能耗；
    获得所述空调系统在所述测试用控制参数下运行时的实际能耗；
    根据公式：能耗比＝实际能耗/空调系统基准能耗，计算得到所述能耗比。
  7. 一种控制参数确定方法，其特征在于，包括：
    将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
    将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
    根据获得的所述控制参数与获得的所述控制偏移量，确定所述空调系统的目标控制参数。
  8. 一种模型训练装置，其特征在于，包括：第一输入单元、第二输入单元、扰动量生成单元、测试参数获得单元、测试记录生成单元和第一训练单元，
    所述第一输入单元，用于将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
    所述第二输入单元，用于将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
    所述扰动量生成单元，用于生成扰动量，根据所述扰动量与所述控制偏移量，确定实际偏移量；
    所述测试参数获得单元，用于根据所述实际偏移量与获得的所述控制参数，获得所述空调系统的一组测试用控制参数；
    所述测试记录生成单元，用于获得所述空调系统在该组所述测试用控制参数下运行时的能耗参数，生成一条测试记录；
    所述第一训练单元，用于根据测试记录获得训练数据，使用所述训练数据对所述当前的偏移量确定模型进行训练，其中，所述训练数据包括：第二环境参数组和实际偏移量。
  9. 一种控制参数确定装置，其特征在于，包括：控制参数获得单元、偏 移量获得单元和目标参数确定单元，
    所述控制参数获得单元，用于将空调系统当前的第一环境参数组输入训练得到的能耗预测模型中，获得在所述能耗预测模型输出的能耗最低时的控制参数，其中，所述能耗预测模型的输入为所述第一环境参数组和所述控制参数，所述能耗预测模型的输出为所述空调系统的能耗；
    所述偏移量获得单元，用于将当前的第二环境参数组输入当前的偏移量确定模型中，获得所述当前的偏移量确定模型输出的控制偏移量，其中，所述偏移量确定模型的输入为所述第二环境参数组，输出为所述控制偏移量；
    所述目标参数确定单元，用于根据获得的所述控制参数与获得的所述控制偏移量，确定所述空调系统的目标控制参数。
  10. 一种存储介质，其特征在于，所述存储介质包括存储的程序，其中，在所述程序运行时控制所述存储介质所在设备执行如权利要求1-6中任一所述的模型训练方法和/或权利要求7所述的控制参数确定方法。
  11. 一种设备，其特征在于，所述设备包括至少一个处理器、以及与所述处理器连接的至少一个存储器、总线；其中，所述处理器、所述存储器通过所述总线完成相互间的通信；所述处理器用于调用所述存储器中的程序指令，以执行如权利要求1-6中任一所述的模型训练方法和/或权利要求7所述的控制参数确定方法。
PCT/CN2020/100520 2019-10-31 2020-07-07 模型训练方法、控制参数确定方法及装置 WO2021082511A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911055197.8A CN112749810B (zh) 2019-10-31 2019-10-31 模型训练方法、控制参数确定方法及装置
CN201911055197.8 2019-10-31

Publications (1)

Publication Number Publication Date
WO2021082511A1 true WO2021082511A1 (zh) 2021-05-06

Family

ID=75644700

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/100520 WO2021082511A1 (zh) 2019-10-31 2020-07-07 模型训练方法、控制参数确定方法及装置

Country Status (2)

Country Link
CN (1) CN112749810B (zh)
WO (1) WO2021082511A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114322382A (zh) * 2021-12-24 2022-04-12 中国电信股份有限公司 机房制冷控制方法、系统、设备及存储介质
CN117691918A (zh) * 2024-02-02 2024-03-12 广东凌霄泵业股份有限公司 一种用于泳池泵电机的控制方法及系统

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113692177B (zh) * 2021-07-09 2023-08-18 厦门科灿信息技术有限公司 数据中心制冷系统功耗的控制方法、装置及终端
CN117387172B (zh) * 2023-12-11 2024-05-17 江苏中江数字建设技术有限公司 基于精准推荐设备控制参数的末端空调节能方法和系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130103214A1 (en) * 2011-10-25 2013-04-25 International Business Machines Corporation Provisioning Aggregate Computational Workloads And Air Conditioning Unit Configurations To Optimize Utility Of Air Conditioning Units And Processing Resources Within A Data Center
CN105571063A (zh) * 2015-12-31 2016-05-11 深圳市同鑫热力技术有限公司 一种浅层地温能能源管理系统及其实现方法
CN107044710A (zh) * 2016-12-26 2017-08-15 深圳达实智能股份有限公司 基于联合智能算法的中央空调节能控制方法及系统
CN109932896A (zh) * 2019-03-22 2019-06-25 易科智控科技(北京)有限公司 一种建筑能耗的控制方法和系统

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108361927A (zh) * 2018-02-08 2018-08-03 广东美的暖通设备有限公司 一种基于机器学习的空调器控制方法、装置以及空调器

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130103214A1 (en) * 2011-10-25 2013-04-25 International Business Machines Corporation Provisioning Aggregate Computational Workloads And Air Conditioning Unit Configurations To Optimize Utility Of Air Conditioning Units And Processing Resources Within A Data Center
CN105571063A (zh) * 2015-12-31 2016-05-11 深圳市同鑫热力技术有限公司 一种浅层地温能能源管理系统及其实现方法
CN107044710A (zh) * 2016-12-26 2017-08-15 深圳达实智能股份有限公司 基于联合智能算法的中央空调节能控制方法及系统
CN109932896A (zh) * 2019-03-22 2019-06-25 易科智控科技(北京)有限公司 一种建筑能耗的控制方法和系统

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114322382A (zh) * 2021-12-24 2022-04-12 中国电信股份有限公司 机房制冷控制方法、系统、设备及存储介质
CN114322382B (zh) * 2021-12-24 2024-03-19 中国电信股份有限公司 机房制冷控制方法、系统、设备及存储介质
CN117691918A (zh) * 2024-02-02 2024-03-12 广东凌霄泵业股份有限公司 一种用于泳池泵电机的控制方法及系统
CN117691918B (zh) * 2024-02-02 2024-04-30 广东凌霄泵业股份有限公司 一种用于泳池泵电机的控制方法及系统

Also Published As

Publication number Publication date
CN112749810B (zh) 2023-08-29
CN112749810A (zh) 2021-05-04

Similar Documents

Publication Publication Date Title
WO2021082511A1 (zh) 模型训练方法、控制参数确定方法及装置
WO2021063033A1 (zh) 空调能耗模型训练方法与空调系统控制方法
WO2020199648A1 (zh) 空调的控制方法和装置
US9429921B2 (en) Method and system for energy control management
WO2021082478A1 (zh) 空调系统能耗预测方法及装置
CN109945420B (zh) 基于负荷预测的空调控制方法、装置及计算机存储介质
WO2023030522A1 (zh) 一种数据中心空调系统诊断方法及装置
CN107036238B (zh) 动态预测外气与负载智慧节能控制方法
CN110631212B (zh) 一种中央空调冷却水系统的节能控制方法
WO2023040694A1 (zh) 制冷机房能效计算方法、装置和电子设备
US20200217544A1 (en) Electronic device and control method thereof
CN111582604A (zh) 数据处理方法和装置、电子设备、及介质
CN110542178B (zh) 具有自学习能力的空调冷冻机房控制方法及系统
Wang et al. A machine learning-based control strategy for improved performance of HVAC systems in providing large capacity of frequency regulation service
CN110135655A (zh) 一种用于确定能源站的运行控制策略的方法与设备
Corbin et al. Transactive control of commercial building hvac systems
CN110762739B (zh) 数据中心空调控制方法、装置、设备及存储介质
CN115906467A (zh) 基于换电站的数据处理方法、装置、电子设备及存储介质
CN115585541A (zh) 空调机房系统的控制方法及装置、电子设备及存储介质
WO2022110213A1 (zh) 生成分析对象的预测模型的方法、装置和存储介质
CN113739368A (zh) 一种中央空调系统的冷站控制方法及系统
CN112413862A (zh) 能耗系数的确定方法和装置、存储介质、电子装置
CN114484724B (zh) 空调控制方法、装置、空调、存储介质和计算机程序产品
CN113068374B (zh) 换热系统的控制方法、装置、设备以及存储介质
CN115877714A (zh) 一种制冷系统控制方法、装置、电子设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20883613

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20883613

Country of ref document: EP

Kind code of ref document: A1