Hot user room temperature measurement and control method based on bluetooth communication and transfer learning

Abstract

The application discloses a kind of hot user room temperature measurement and control method based on bluetooth communication and transfer learning, comprising: selecting multiple typical users as room temperature measuring point, and selecting effective fixed position to install room temperature collection device in turn;The room temperature collection device at least includes temperature collection module, bluetooth communication module and control module;Room temperature collection device is transmitted to user bluetooth smart terminal after establishing bluetooth communication connection with user bluetooth smart terminal by its bluetooth communication module, and the indoor temperature collected;After comparing and analyzing indoor temperature with the target temperature value set, the heat network balance optimization scheduling model with hot user indoor temperature as control target is established using convolutional neural network and transfer learning method;After the valve and water pump are feedforward regulated by heat network balance optimization scheduling model, the indoor temperature collected is acquired again, and whether indoor temperature meets the standard is judged.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent heating room temperature measurement, and particularly relates to a heating user room temperature measurement and control method based on Bluetooth communication and transfer learning. BACKGROUND

[0002] In the heating industry, intelligent heating has been formally proposed and widely focused by people with insight in the industry. In intelligent heating, indoor temperature collection of heating users is particularly important. Room temperature monitoring data is not only the basis for heating effect, but also the data basis for the whole network control strategy of intelligent heating. In the intelligent heating system, whether it is the demand side of the heating user or the supply side of the heating enterprise, the representative key parameter of heating is the indoor temperature of the heating user. Therefore, indoor temperature monitoring is an important means to realize intelligent heating, which not only feeds back the real-time indoor heating temperature of the heating user, but also serves as the basis for judging the heating state of the heating user and the guarantee effect of the heating system.

[0003] However, there is an imbalance in the amount of heat in the heating pipe network, and the phenomenon of near heat and far cold often occurs. The indoor temperature of the household does not meet the standard, and complaints from heating users occur from time to time. Among them, the indoor temperature information of the terminal heating user is the most direct parameter representing the heating effect. The user indoor temperature information can be collected in time, and the heating effect can be intuitively understood. The water supply temperature and flow can be adjusted in time according to the actual situation, which can effectively reduce the occurrence of customer complaints and further prevent energy waste caused by over-supply.

[0004] However, at present, the collection method of indoor temperature of terminal users of many heating companies is basically in a passive situation, that is, when a customer complaint occurs, the temperature is measured by artificial temperature measurement in the user's home. This results in the problems of untimely and discontinuous data collection, which greatly affects the quality of service and the timeliness of problem solving. When wireless room temperature collection is performed by using GPRS, NB-IOT, and LORA technology, the cost is high, and users and heating enterprises are not willing to accept it.

[0005] In addition, for a new heating station, there are few operation data, and it is difficult to establish a heat network balance scheduling model based on collected indoor temperature and historical operation data. In addition, for the same heating station, the heating scene will change compared with the historical heating season when entering a new heating season. The heat network balance scheduling model based on historical heating season data will have a phenomenon of decreased model precision when used in the new heating season.

[0006] Based on the above technical problems, a new heating user room temperature measurement and control method based on Bluetooth communication and transfer learning is needed. SUMMARY

[0007] The application aims to provide a heat user room temperature measurement and control method based on Bluetooth communication and transfer learning, which can effectively collect the indoor temperature of users in time, has the characteristics of low cost and low power consumption based on Bluetooth module, and can transfer the heat network balance scheduling model of the source domain to the target domain by using the transfer learning algorithm according to the collected indoor user temperature, the operation data of the heating system and the weather data, thereby avoiding the problems of less data of the new heating station and difficult collection and analysis, and the problem of decreased accuracy of the heat network balance scheduling model in the new heating season.

[0008] To solve the above technical problems, the application provides a heat user room temperature measurement and control method based on Bluetooth communication and transfer learning, which is characterized by comprising the following steps:

[0009] Step S1, a plurality of typical users are selected as room temperature measurement points, and room temperature collection devices are installed in effective fixed positions in turn; the room temperature collection device at least comprises a temperature collection module, a Bluetooth communication module and a control module;

[0010] Step S2, the room temperature collection device transmits the collected indoor temperature to the user Bluetooth intelligent terminal after establishing a Bluetooth communication connection with the user Bluetooth intelligent terminal through the Bluetooth communication module;

[0011] Step S3, the background server compares and analyzes the indoor temperature with the set target temperature value, and establishes a heat network balance optimization scheduling model with the indoor temperature of the heat user as the control target by using the convolutional neural network and the transfer learning method;

[0012] Step S4, the indoor temperature collected is acquired again after the valve and the water pump are feedforward regulated through the heat network balance optimization scheduling model, and whether the indoor temperature meets the standard is judged.

[0013] Further, in step S1, the plurality of typical users are selected as room temperature measurement points, and the room temperature collection devices are installed in effective fixed positions in turn, and specifically comprising:

[0014] Different buildings with long, medium and short distances from the heating station are selected as typical buildings for room temperature collection;

[0015] The bottom, middle and top layers of different buildings are selected as typical users as room temperature measurement points;

[0016] One or more effective fixed positions are selected for installing the room temperature collection devices according to the room area of different users; the effective fixed position is: the distance from the inner surface of the outer wall is not less than (1.5±0.05) m, the distance from the inner wall surface is not less than (1.0±0.05) m, and the distance from the ground directly above is (1.4±0.05) m;

[0017] And after installing the room temperature collection device, further comprising: through the administrator Bluetooth smart terminal to obtain and record the effective position information of the user Bluetooth smart terminal and the data information of the room temperature collection device in each typical user, at least including building identification, household identification, collection device identification, current indoor temperature and recording time.

[0018] Further, in the step S2, the room temperature collection device transmits the collected indoor temperature to the user Bluetooth smart terminal after establishing a Bluetooth communication connection with the user Bluetooth smart terminal through its Bluetooth communication module, specifically including:

[0019] After turning on the Bluetooth function of the user Bluetooth smart terminal, the room temperature collection client of the user smart terminal will automatically search for available Bluetooth devices, i.e. room temperature collection devices, within the range of the Bluetooth module, and after matching the device name and password, it will automatically establish a Bluetooth communication connection with the data collection device and stop searching for devices.

[0020] After the temperature collection module in the data collection device collects the indoor temperature, it transmits a data packet to the user Bluetooth smart terminal through the Bluetooth communication module;

[0021] After the user Bluetooth smart terminal receives the data packet, it obtains the basic information of the data packet, including indoor temperature information, collection device identification and current time;

[0022] Wherein, when establishing a Bluetooth communication connection, it is determined whether the position information of the user Bluetooth smart terminal is in the calibrated preset position, if it is in the preset position, the Bluetooth communication connection is allowed to be established; otherwise, the Bluetooth communication connection is prohibited.

[0023] Further, in the step S3, after comparing and analyzing the indoor temperature with the set target temperature value, the background server establishes a heat network balance optimization scheduling model with the indoor temperature of the heat user as the control target using convolutional neural network and transfer learning method, specifically including:

[0024] The background server calculates the deviation of the indoor temperature from the set target temperature value, and determines whether the deviation is within the preset range, if the deviation is not within the preset range, a heat network balance optimization scheduling model with the heat user as the control target is established using convolutional neural network and transfer learning method, i.e.

[0025] Generate a heat network balance scheduling model by training historical source domain data using convolutional neural network and back propagation algorithm;

[0026] The heat network balancing scheduling model is transferred and fine-tuned: the model parameters and weights of the heat network balancing scheduling model trained in the source domain are loaded, except for the last fully connected layer. A complete fully connected model is retrained using the target domain data, and the parameters and weights extracted from the last fully connected model are saved as the feature extraction source in the next stage of the fully connected layer. The training features and weights of the source domain model in the previous stage are loaded into the fully connected layer, the training parameters of some layers or the entire model layer structure are frozen, and the heat network balancing scheduling model after transfer training is fine-tuned using the source domain data and the backpropagation algorithm.

[0027] The target domain data is input into the fine-tuned heat network balance scheduling model, and then the predicted value of the water supply temperature is output. The water supply temperature is then adjusted to ensure that the water supply temperature reaches the predicted value, and the indoor temperature of the heat users is controlled by adjusting the water supply temperature.

[0028] Furthermore, a heat network balancing and scheduling model is trained using historical source domain data, specifically including:

[0029] Historical operating data at time N, weather data at time N, and indoor temperature data at time NT are obtained from the heating system as source domain input variables, and indoor temperature at time N+T is used as source domain output variable; the historical operating data includes supply water temperature, return water temperature, supply water pressure, return water pressure, and instantaneous flow rate; the weather data includes outdoor temperature data, humidity data, wind speed, and external air pressure;

[0030] The acquired source domain data undergoes preprocessing: deduplication, missing data completion, and outlier handling. For deduplication, duplicate data stored at the same time are removed, retaining only one instance. For data completion, if data exists at both ends of the sequence, the average of adjacent data is used to replace missing values; if data is missing at the beginning or end of the sequence, the average of data from the previous and next week at that time is used. For outlier handling, the 3-sigma method is used to identify outliers in the data sequence, and detected outliers are replaced with adjacent values.

[0031] The preprocessed source domain data is divided into a 70% training dataset and a 30% test dataset.

[0032] After optimizing the convolutional feature mapping network unit and the fully connected classification network unit in the convolutional neural network model through the backpropagation algorithm, the source domain training dataset is input into the optimized convolutional neural network model for training to generate a heat network balancing scheduling model.

[0033] The accuracy of the trained heat network balancing scheduling model is calculated using the source domain test dataset. If the accuracy is less than the preset accuracy, the convolutional neural network model is re-optimized; otherwise, training is stopped and the parameters of the heat network balancing scheduling model are saved.

[0034] Further, the fine-tuning of the heat network balance scheduling model after migration training using source domain data and a back propagation algorithm specifically includes:

[0035] Extracting m-sized micro-call training data from the labeled source domain training data set;

[0036] Using a back propagation algorithm to optimize the convolution feature mapping network unit and the fully connected network unit in the heat network balance scheduling model;

[0037] By iteratively optimizing the loss function, the parameters of the heat network balance scheduling model are constrained and the objective function is set to continuously train the model until the maximum number of iterations is reached or the loss function is less than the set value, the objective function reaches the minimum value, the training is stopped, the obtained model parameters are saved, and the migration training and fine-tuning of the heat network balance scheduling model are completed.

[0038] Further, when the heat network balance scheduling model is migrated and trained, a domain adaptation module is set in the heat network balance scheduling model, and the domain adaptation module uses three kinds of transfer learning algorithms, TCA, JDA and GFK, to map from source domain data to target domain data, increasing the similarity between source domain data and target domain data.

[0039] Further, the convolutional neural network further includes a feature extraction module for extracting features from the new source domain and target domain data after domain adaptation. The feature extraction module includes three convolutional networks, each including a convolutional layer, a pooling layer and a Relu activation layer. The first convolutional network includes 32 convolutional kernels with a size of [3, 3, 1, 32] and a step size of 1. The pooling layer uses 2*2 max pooling, and the input data is reduced to [16*16] after passing through the first convolutional network. The second convolutional network includes 64 convolutional kernels with a size of [3, 3, 32, 64] and a step size of 1. The pooling layer uses 2*2 max pooling, and the output size after passing through the second convolutional network is [8*8]. The third convolutional network includes 128 convolutional kernels with a size of [3, 3, 64, 128] and a step size of 1. The pooling layer uses 2*2 max pooling, and the output size after passing through the second convolutional network is [4*4]. Then there is a fully connected layer with a size of [4*4*128, 512] for feature integration.

[0040] Further, the convolutional neural network includes an input layer, a convolutional layer, an activation layer, a pooling layer, a fully connected layer and a final output layer.

[0041] Further, the heat network balance optimization scheduling model is used to perform feedforward regulation on the valve and the water pump, including:

[0042] The lag time of the heat supply system is approximately equal to the time of fluid flowing in the pipeline, the fluid flow time is selected as the feedforward regulation time of the heat supply system, and the valve and the water pump of the heat supply system are adjusted in advance;

[0043] The feedforward regulation time calculation is expressed as: L is the pipeline length, v w is the flow rate of the fluid in the pipeline, D w is the diameter of the pipeline, G w is the flow of the fluid in the pipeline.

[0044] The beneficial effects of the present application are:

[0045] (1) The present application selects multiple typical users as room temperature measurement points, and selects effective fixed positions to install room temperature collection devices in turn; the room temperature collection device transmits the collected indoor temperature to the user's Bluetooth smart terminal after establishing a Bluetooth communication connection with the user's Bluetooth smart terminal through its Bluetooth communication module; the background server compares and analyzes the indoor temperature with the set target temperature value, and establishes a heat network balance optimization scheduling model with the indoor temperature of the hot user as the control target by using the convolutional neural network and the transfer learning method; the indoor temperature collected is obtained again after the valve and the water pump are feedforward regulated by the heat network balance optimization scheduling model, and it is judged whether the indoor temperature meets the standard; the indoor temperature of the user can be collected in time and effectively, and it is not necessary to frequently go to the user's home to measure the room temperature, and the collection based on the Bluetooth module has the characteristics of low cost and low power consumption, and the transfer learning algorithm can transfer the heat network balance scheduling model of the source domain to the target domain according to the collected indoor user temperature combined with the heat supply system operation data and weather data, avoiding the problems of insufficient data of the new heat station and difficult collection and analysis, and the problem of decreased precision of the heat network balance scheduling model in the new heating season;

[0046] (2) The present application maps the source domain data to the target domain data through the domain adaptation module, increases the similarity of the source domain data and the target domain data; the feature extraction module extracts the features of the new source domain and the target domain data after domain adaptation; the model loss can be optimized, the domain-invariant features can be quickly obtained, the model transfer efficiency can be improved, and the model precision can be ensured; and the transfer model is fine-tuned by using the back propagation algorithm, avoiding the problem of over-transfer;

[0047] (3) The transfer learning based on the convolutional neural network can not only be used to solve the small sample problem of insufficient training data, but also can further improve the performance and robustness of the convolutional neural network by utilizing the model experience of the existing system of the source domain;

[0048] (4) The present application considers that the adjustment response of indoor temperature to water supply temperature has a certain time lag and attenuation, therefore, the feedforward regulation time is calculated, and the regulation is performed in advance to ensure that the indoor temperature can quickly respond to the standard.

[0049] Other features and advantages of the present application will be set forth in the descriptions that follow, and in part will be apparent from the description, or can be learned by practice of the application. The purposes and other advantages of the present application will be realized and attained by the structures particularly pointed out in the description and the appended drawings.

[0050] In order to make the above objectives, features and advantages of the present application more apparent, the following will describe a preferred embodiment in detail, and the accompanying drawings will be described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0052] Figure 1 is a flow chart of the hot user room temperature measurement and control method based on Bluetooth communication and transfer learning.

[0053] Figure 2 is a principle diagram of the hot user room temperature measurement and control method based on Bluetooth communication and transfer learning.

[0054] Figure 3 is a transfer learning framework diagram based on convolutional neural network. DETAILED DESCRIPTION

[0055] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more apparent, the technical solutions of the present application will be described clearly and completely in the following with reference to the drawings. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the present application.

[0056] Embodiment 1

[0057] Figure 1 is a flow chart of the hot user room temperature measurement and control method based on Bluetooth communication and transfer learning.

[0058] Figure 2 is a principle diagram of the hot user room temperature measurement and control method based on Bluetooth communication and transfer learning.

[0059] As Figure 1As shown, the embodiment 1 provides a heat user room temperature measurement and control method based on Bluetooth communication and transfer learning, characterized in that the heat user room temperature measurement and control method comprises:

[0060] Step S1, select a plurality of typical users as room temperature measurement points, and select effective fixed positions to install room temperature collection devices in turn; the room temperature collection device at least includes a temperature collection module, a Bluetooth communication module and a control module;

[0061] Step S2, the room temperature collection device transmits the collected indoor temperature to the user Bluetooth smart terminal after establishing a Bluetooth communication connection with the user Bluetooth smart terminal through the Bluetooth communication module thereof;

[0062] Step S3, the background server compares and analyzes the indoor temperature with the set target temperature value, and establishes a heat network balance optimization scheduling model with the heat user indoor temperature as the control target by using the convolutional neural network and the transfer learning method;

[0063] Step S4, the indoor temperature collected is acquired again after the valve and the water pump are feedforward regulated through the heat network balance optimization scheduling model, and it is judged whether the indoor temperature meets the standard.

[0064] In this embodiment, in step S1, a plurality of typical users are selected as room temperature measurement points, and effective fixed positions are selected to install room temperature collection devices in turn, which specifically includes:

[0065] Different buildings with long, medium and short distances from the heat supply station are selected as typical buildings for room temperature collection;

[0066] The bottom, middle and top residents of different buildings are selected as typical users as room temperature measurement points;

[0067] One or more effective fixed positions are selected for installing room temperature collection devices according to the room area of different users; the effective fixed position is: the distance from the inner surface of the outer wall is not less than (1.5±0.05)m, the distance from the inner wall surface is not less than (1.0±0.05)m, and the distance from the ground directly above is (1.4±0.05)m;

[0068] And after installing the room temperature collection device, it further includes: obtaining and recording the effective position information of the user Bluetooth smart terminal and the data information of the room temperature collection device of each typical user through the administrator Bluetooth smart terminal, including at least building identification, resident identification, collection device identification, current indoor temperature and recording time.

[0069] In this embodiment, in step S2, the room temperature collection device transmits the collected indoor temperature to the user Bluetooth smart terminal after establishing a Bluetooth communication connection with the user Bluetooth smart terminal through the Bluetooth communication module thereof, which specifically includes:

[0070] After the Bluetooth function of the user Bluetooth smart terminal is turned on, the room temperature acquisition client of the user smart terminal will automatically search for the available Bluetooth devices in the Bluetooth module range, that is, the room temperature acquisition device, and automatically establish a Bluetooth communication connection with the data acquisition device after matching the device name and password of the device, and stop searching for the device.

[0071] After the temperature acquisition module in the data acquisition device acquires the indoor temperature, the data packet is transmitted to the user Bluetooth smart terminal through the Bluetooth communication module.

[0072] After the user Bluetooth smart terminal receives the data packet, the basic information of the data packet is obtained, including the indoor temperature information, the acquisition device identifier, and the current time.

[0073] In the process of establishing the Bluetooth communication connection, it is determined whether the position information of the user Bluetooth smart terminal is in the calibrated preset position, and if the preset position is allowed to establish the Bluetooth communication connection, otherwise the Bluetooth communication connection is prohibited.

[0074] It should be noted that the user Bluetooth smart terminal can be in various forms, at least including a mobile phone, a set-top box, and a router, so that the indoor temperature of the user can be conveniently and low-costly acquired through the smart terminal commonly used by the user at home; in order to ensure the privacy of the user and prevent the information from being stolen by illegal persons, the indoor temperature and user attribute information to be transmitted can be encrypted, the user is set with an authorized access mechanism, so that the transmitted information is in cipher text, rather than in plain text, and only the authorized user access is authorized, thereby effectively ensuring the privacy of the user and the security of the information; and the position information of the user Bluetooth smart terminal is calibrated, and only within the effective position range, the Bluetooth communication connection can be established to acquire the indoor temperature data of the user, thereby preventing other malicious smart terminals from acquiring the indoor temperature data of the user.

[0075] Figure 3 It is a migration learning framework based on a convolutional neural network.

[0076] As shown in Figure 3 In this embodiment, in step S3, the background server compares and analyzes the indoor temperature with the set target temperature value, and establishes a heat network balance optimization scheduling model with the heat user indoor temperature as the control target by using the convolutional neural network and the migration learning method, which specifically includes:

[0077] The background server calculates the deviation of the indoor temperature from the set target temperature value, and determines whether the deviation is within the preset range, and if the deviation is not within the preset range, the heat network balance optimization scheduling model with the heat user as the control target is established by using the convolutional neural network and the migration learning method, that is:

[0078] The heat network balance scheduling model is trained by a convolutional neural network and a back propagation algorithm;

[0079] The heat network balance scheduling model is trained by a convolutional neural network and a back propagation algorithm;

[0080] The target domain data is input into the fine-tuned heat network balance scheduling model, and then the predicted value of the water supply temperature is output, and the water supply temperature is adjusted to the predicted value by adjusting the valve opening, and the indoor temperature of the heat user is controlled by adjusting the water supply temperature.

[0081] In actual application, the heat network balance scheduling model of heat station A which has been running for many years can be applied to heat station B which has just been put into operation by using the method of transfer learning. Since heat station B which has just been put into operation has less historical operation, it is difficult to effectively establish a scheduling model, therefore, the source domain information of heat station A needs to be fully utilized to adjust the network structure of the scheduling model based on a small amount of target data of heat station B, so that the heat network balance scheduling model after transfer learning can be effectively applied in heat station B. Or for the same heat station, the heating scene will change compared with the historical heating season when entering a new heating season, and the model precision will decrease when the heat network balance scheduling model established based on the historical heating season data is used in the new heating season, so the model structure and parameters need to be adjusted by forward learning.

[0082] It should be noted that the back propagation stage uses a back propagation algorithm combined with an optimization method, which corrects the error signal by using an optimization algorithm while performing error propagation, and fine-tunes the network model. This method calculates the gradient of all loss functions in the network, and then feeds back this gradient to the optimization method to update the weight values in order to obtain the minimized loss function or cost function. That is, the gradient is iteratively calculated for each layer of neural nodes using the chain rule during the back propagation process, and the error between the actual output and the corresponding ideal output is transmitted.

[0083] In actual application, different levels of control strategies are set to guide the on-demand heating, including: heat station adjustment, building heat inlet adjustment valve adjustment and household adjustment valve adjustment; the three adjustment modes are preset with priority levels, and adjustment is carried out in order of levels to ensure that the indoor temperature of the user meets the standard; when the indoor temperature meets the standard after adjustment according to the adjustment mode with a higher priority level, adjustment of the control strategy of the subsequent level is not carried out; when the indoor temperature still does not meet the standard after adjustment according to the adjustment mode with a higher priority level, adjustment of the control strategy of the next level is carried out in turn, and if the indoor temperature still does not meet the standard, manual on-site processing is required. The heat station adjustment content is to create a control strategy for the heat station with the indoor temperature as the target, and the adjustment content includes secondary pipe network water supply temperature, flow and the like; the building heat inlet adjustment valve adjustment is to adjust the heat inlet flow with the heat inlet supply and return water temperature difference of each building as the target, so that the indoor temperature meets the standard; compared with the building heat inlet adjustment valve adjustment mode, the inlet adjustment valve adjustment mode differs in that the adjustment device is changed from the building heat inlet electric adjustment valve to the household electric adjustment valve, so that the indoor temperature meets the standard.

[0084] After the entire network structure is defined, the back propagation algorithm is used to train the entire network parameters, and the convolutional neural network obtains the best model for prediction through the training sample data set. In order to measure the prediction effect, the target function C is defined as the judgment of the prediction result:

[0085] Wherein, y(x) is the label of the sample, and aL(x) is the output result value. The smaller C is, the better the training effect is. In this process, the gradient descent algorithm is often used to obtain the parameters with the minimum target function C. The core idea of the gradient descent algorithm is to take the partial derivative of each parameter, change the parameter in the direction that makes C smaller, and iterate until C reaches the minimum value.

[0086] When the training data is insufficient, overfitting problem may occur. In order to obtain a better heat network balance scheduling model, sufficient data features are often injected during the training process, and a large-scale data set is required to provide sufficient source data. The Dropout technology can prevent overfitting and is applied in the fully connected layer of the convolutional neural network. This technology randomly makes some hidden layer neurons in the network not work and discards the connectivity between neurons by setting the output of neurons to 0 according to a certain probability during model training, thereby effectively preventing overfitting and reducing the coupling between different parameters; by introducing the Dropout layer, the overfitting problem is avoided, the precision of the user heat network balance scheduling model is improved, and a transfer learning model is established to further improve the model precision and reduce the network training time, thereby improving the efficiency of establishing the heat network balance scheduling model.

[0087] In addition, the anti-overfitting performance and model accuracy of the convolutional neural network can be improved by introducing data enhancement transformation, modifying the structure of the convolutional neural network, adjusting the size of the convolution kernel and the pooling unit, using a gradient descent training optimization algorithm with an automatically updated learning rate, and a more intense Dropout dropout rate, according to the target data domain problem.

[0088] In this embodiment, the heat network balance scheduling model is trained using historical source domain data, specifically including:

[0089] The N-time historical operation data, N-time weather data, N-T-time indoor temperature data, and N-time indoor temperature data are obtained from the heating system as source domain input variables, and the N+T-time indoor temperature is obtained as a source domain output variable. The historical operation data includes water supply temperature, return water temperature, water supply pressure, return water pressure, and instantaneous flow; the weather data includes outdoor temperature data, humidity data, wind speed, and external air pressure;

[0090] The obtained source domain data is preprocessed: removing duplicate data, filling missing data, and handling outliers; when removing duplicate data, duplicate data stored at the same time is removed, and one is retained; when filling data, if there are data at both ends of the time, the average value of the adjacent two data is used to replace the missing value, if the sequence is missing data at the beginning or end, the average value of the data at that time of the previous week and the next week is used to replace the missing value; when handling outliers, the 3-sigma method is used to detect outliers in the data sequence, and the adjacent value is used to replace the detected outliers;

[0091] The preprocessed source domain data is divided into 70% training data set and 30% test data set;

[0092] After optimizing the convolution feature mapping network unit and the fully connected classification network unit in the convolutional neural network model through the back propagation algorithm, the source domain training data set is input into the optimized convolutional neural network model to generate a heat network balance scheduling model;

[0093] The accuracy of the heat network balance scheduling model generated by training is calculated through the source domain test data set. If the accuracy is less than the preset accuracy, the convolutional neural network model is re-optimized; otherwise, the training is stopped, and the parameters of the heat network balance scheduling model are saved.

[0094] In this embodiment, the source domain data and the back propagation algorithm are used to fine-tune the heat network balance scheduling model after migration training, specifically including:

[0095] The fine-tuning training data of size m is extracted from the labeled source domain training data set;

[0096] The convolution feature mapping network unit and the full connection network unit in the heat network balance scheduling model are optimized using a back propagation algorithm;

[0097] The parameters of the heat network balance scheduling model are constrained and set by iteratively optimizing the loss function, and the model is continuously trained until the maximum iteration number is reached or the loss function is less than the set value, the target function reaches the minimum value, the training is stopped, the obtained model parameters are saved, and the migration training and fine-tuning of the heat network balance scheduling model are completed.

[0098] In this embodiment, when the heat network balance scheduling model is migrated and trained, a domain adaptation module is provided in the heat network balance scheduling model, and the domain adaptation module uses three kinds of transfer learning algorithms, TCA, JDA and GFK, to map from source domain data to target domain data, and increase the similarity between source domain data and target domain data.

[0099] In this embodiment, the convolutional neural network further includes a feature extraction module for extracting features from the new source domain and target domain data after domain adaptation. The feature extraction module includes three convolutional networks, each of which includes a convolutional layer, a pooling layer and a Relu activation layer. In the first convolutional network, the number of convolutional kernels is 32, the size is [3, 3, 1, 32], the step is 1, the pooling layer uses 2*2 maximum pooling, and the input data is changed to [16*16] after passing through the first convolutional network. In the second convolutional network, the number of convolutional kernels is 64, the size is [3, 3, 32, 64], the step is 1, the pooling layer uses 2*2 maximum pooling, and the output size after passing through the second convolutional network is [8*8]. In the third convolutional network, the number of convolutional kernels is 128, the size is [3, 3, 64, 128], the step is 1, the pooling layer uses 2*2 maximum pooling, and the output size after passing through the second convolutional network is [4*4]. Then there is a full connection layer with a size of [4*4*128, 512] for feature integration.

[0100] In this embodiment, the convolutional neural network includes an input layer, a convolutional layer, an activation layer, a pooling layer, a full connection layer and a final output layer.

[0101] In this embodiment, the valve and the water pump are adjusted by the heat network balance optimization scheduling model, including:

[0102] The lag time of the heating system is approximately equal to the time of fluid flowing in the pipeline, and the fluid flow time is selected as the feedforward adjustment time of the heating system, and the valve and the water pump of the heating system are adjusted in advance;

[0103] The feedforward adjustment time calculation is expressed as: L is the length of the pipeline, v w is the flow rate of the fluid in the pipeline, D wDiameter of the pipe, G w Flow rate of the fluid in the pipe.

[0104] It should be noted that the indoor temperature is mainly affected by the building maintenance structure, the orientation of the building, the outdoor temperature, etc., and the indoor temperature has a certain time lag and attenuation in the adjustment response to the supply water temperature, so a feedforward dynamic adjustment model needs to be established to obtain the feedforward adjustment time.

[0105] The present application selects multiple typical users as room temperature measurement points, and selects effective fixed positions to install room temperature collection devices in turn; the room temperature collection device transmits the collected indoor temperature to the user's Bluetooth smart terminal after establishing a Bluetooth communication connection with the user's Bluetooth smart terminal through its Bluetooth communication module; the background server compares and analyzes the indoor temperature with the set target temperature value, and establishes a heat network balance optimization scheduling model with the indoor temperature of the hot user as the control target by using the convolutional neural network and the transfer learning method; the valve and the water pump are adjusted by the heat network balance optimization scheduling model, and the collected indoor temperature is obtained again to determine whether the indoor temperature meets the standard; the indoor temperature of the user can be collected in time and effectively, and it is not necessary to frequently go to the user's home to measure the room temperature, and the collection based on the Bluetooth module has the characteristics of low cost and low power consumption, and according to the collected indoor user temperature, the heat supply system operation data and weather data, the transfer learning algorithm can be used to migrate the heat network balance scheduling model of the source domain to the target domain, avoiding the problems of insufficient data of the new heat station and difficult collection and analysis, and the problem of decreased precision of the heat network balance scheduling model in the new heating season.

[0106] The present application maps the source domain data to the target domain data through the domain adaptation module, increases the similarity of the source domain data and the target domain data; through the feature extraction module, the new source domain and target domain data after domain adaptation are subjected to feature extraction; the model loss can be optimized, the domain invariant features can be quickly obtained, the model transfer efficiency can be improved, and the model precision can be ensured; and the transfer model is fine-tuned by using the back propagation algorithm, and the over-transfer problem is avoided.

[0107] The present application uses the transfer learning based on the convolutional neural network, which can not only solve the small sample problem caused by insufficient training data, but also further improve the performance and robustness of the convolutional neural network by using the model experience of the existing system of the source domain.

[0108] The present application considers that the indoor temperature has a certain time lag and attenuation in the adjustment response to the supply water temperature, so the feedforward adjustment time is calculated to adjust in advance, so as to ensure that the indoor temperature can quickly respond to the standard.

[0109] In several embodiments provided in the present application, it should be understood that the disclosed system and method can also be implemented in other manners. The above described system embodiments are merely exemplary. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architectures, functions and operation of the system, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a segment or a portion of code which comprises one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions shown in the blocks can occur in different orders than those shown in the figures. For example, two blocks shown in succession can in fact be executed substantially concurrently or in reverse order, depending on the functionality involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented by dedicated hardware-based systems which perform the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0110] In addition, the various functional modules in the embodiments of the present application can be integrated together or exist separately, or two or more modules can be integrated to form an independent part.

[0111] If the functions are implemented in the form of software function modules and sold or used as an independent product, they can be stored in a computer readable storage medium. Based on such an understanding, the technical solutions of the present application essentially or partially, or a part of the technical solutions can be embodied in a software product, and the computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods according to the embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various media which can store program codes.

[0112] With the above ideal embodiments according to the present application as the inspiration, through the above description, relevant personnel can make various changes and modifications without deviating from the scope of the technical idea of the present application. The technical scope of the present application is not limited to the content in the specification, and must be determined according to the scope of the claims.

Claims

1. A method for measuring and controlling the room temperature of hot users based on Bluetooth communication and transfer learning, characterized in that, The method for measuring and controlling the room temperature of heat users includes: Step S1: Select multiple typical users as room temperature measurement points, and install room temperature acquisition devices in sequence at valid fixed locations; the room temperature acquisition device includes at least a temperature acquisition module, a Bluetooth communication module, and a control module; Step S2: After the room temperature acquisition device establishes a Bluetooth communication connection with the user's Bluetooth smart terminal through its Bluetooth communication module, it transmits the acquired indoor temperature to the user's Bluetooth smart terminal. Step S3: After comparing and analyzing the indoor temperature with the set target temperature value, the backend server uses convolutional neural networks and transfer learning methods to establish a heat network balance optimization scheduling model with the indoor temperature of heat users as the control target. Step S4: After adjusting the valves and water pumps using the heat network balance optimization scheduling model, the indoor temperature is collected again to determine whether the indoor temperature meets the standard. In step S3, the backend server compares and analyzes the indoor temperature with the set target temperature value, and then uses convolutional neural networks and transfer learning methods to establish a heat network balance optimization scheduling model with the indoor temperature of heat users as the control target. Specifically, this includes: The backend server calculates the deviation between the indoor temperature and the set target temperature value, and determines whether the deviation is within a preset range. If the deviation is not within the preset range, a heat network balance optimization scheduling model with heat users as the control target is established using convolutional neural networks and transfer learning methods. Specifically, this includes: A heat network balancing scheduling model is generated by training historical source domain data using a convolutional neural network and backpropagation algorithm. The heat network balancing scheduling model is transferred and fine-tuned: the model parameters and weights of the heat network balancing scheduling model trained in the source domain are loaded, except for the last fully connected layer. A complete fully connected model is retrained using the target domain data, and the parameters and weights extracted from the last fully connected model are saved as the feature extraction source in the next stage of the fully connected layer. The training features and weights of the source domain model in the previous stage are loaded into the fully connected layer, the training parameters of some layers or the entire model layer structure are frozen, and the heat network balancing scheduling model after transfer training is fine-tuned using the source domain data and the backpropagation algorithm. The target domain data is input into the fine-tuned heat network balance scheduling model, and then the predicted value of the water supply temperature is output. The water supply temperature is then adjusted to ensure that the water supply temperature reaches the predicted value, and the indoor temperature of the heat users is controlled by adjusting the water supply temperature.

2. The method for measuring and controlling room temperature of heat users according to claim 1, characterized in that, In step S1, selecting multiple typical users as room temperature measurement points and sequentially installing room temperature acquisition devices at effective fixed locations specifically includes: Different buildings at distances of far, medium, and near relative to the heating station were selected as typical buildings for room temperature data collection. For different buildings, the ground floor, middle floor and top floor residents were selected as typical users as the temperature measurement points; For different users' room areas, select one or more effective fixed locations to install room temperature monitoring devices; the effective fixed locations are: the room temperature monitoring device is not less than (1.5±0.05)m from the inner surface of the outer wall, the room temperature monitoring device is not less than (1.0±0.05)m from the inner surface of the inner wall, and the room temperature monitoring device is (1.4±0.05)m directly above the ground; In addition to installing the room temperature acquisition device, it also includes: obtaining and recording the effective location information of the user's Bluetooth smart terminal and the data information of the room temperature acquisition device in each typical user through the administrator's Bluetooth smart terminal, including at least the building identifier, resident identifier, acquisition device identifier, current indoor temperature and recording time.

3. The method for measuring and controlling room temperature of heat users according to claim 1, characterized in that, In step S2, after the room temperature acquisition device establishes a Bluetooth communication connection with the user's Bluetooth smart terminal through its Bluetooth communication module, it transmits the acquired indoor temperature to the user's Bluetooth smart terminal, specifically including: After enabling the Bluetooth function of the user's Bluetooth smart terminal, the user's smart terminal's room temperature acquisition client will automatically search for available room temperature acquisition devices within the Bluetooth module range, match their device names and passwords, and then automatically establish a successful Bluetooth communication connection with the data acquisition device and stop searching for devices. After the temperature acquisition module in the data acquisition device acquires the indoor temperature, it transmits data packets to the user's Bluetooth smart terminal through the Bluetooth communication module. After receiving the data packet, the user's Bluetooth smart terminal obtains the basic information of the data packet, including indoor temperature information, the identification of the data acquisition device, and the current time. Specifically, when establishing a Bluetooth communication connection, it is determined whether the location information of the user's Bluetooth smart terminal is at a preset location. If it is at the preset location, the establishment of a Bluetooth communication connection is allowed; otherwise, the establishment of a Bluetooth communication connection is prohibited.

4. The method for measuring and controlling room temperature of heat users according to claim 1, characterized in that, Training a heat network balancing scheduling model using the historical source domain data specifically includes: Historical operating data at time N, weather data at time N, and indoor temperature data at time NT are obtained from the heating system as source domain input variables, and indoor temperature at time N+T is used as source domain output variable; the historical operating data includes supply water temperature, return water temperature, supply water pressure, return water pressure, and instantaneous flow rate; the weather data includes outdoor temperature data, humidity data, wind speed, and external air pressure; The acquired source domain data undergoes preprocessing: deduplication, missing data completion, and outlier handling. For deduplication, duplicate data stored at the same time are removed, retaining only one instance. For data completion, if data exists at both ends of the sequence, the average of adjacent data is used to replace missing values; if data is missing at the beginning or end of the sequence, the average of data from the previous and next week at that time is used. For outlier handling, the 3-sigma method is used to identify outliers in the data sequence, and detected outliers are replaced with adjacent values. The preprocessed source domain data is divided into a 70% training dataset and a 30% test dataset. After optimizing the convolutional feature mapping network unit and the fully connected classification network unit in the convolutional neural network model through the backpropagation algorithm, the source domain training dataset is input into the optimized convolutional neural network model for training to generate a heat network balancing scheduling model. The accuracy of the trained heat network balancing scheduling model is calculated using the source domain test dataset. If the accuracy is less than the preset accuracy, the convolutional neural network model is re-optimized; otherwise, training is stopped and the parameters of the heat network balancing scheduling model are saved.

5. The method for measuring and controlling room temperature for heat users according to claim 1, characterized in that, The heat network balancing scheduling model after transfer learning is fine-tuned using the source domain data and backpropagation algorithm, specifically including: Extract micro-call training data of size m from the labeled source domain training dataset; The backpropagation algorithm is used to optimize the convolutional feature mapping network unit and the fully connected network unit in the heat network balancing scheduling model. The parameters of the heating network balancing scheduling model are constrained and the objective function is set by iteratively optimizing the loss function. The model is continuously trained until the maximum number of iterations is reached, or the loss function is less than the set value and the objective function reaches the minimum value. Training is then stopped, and the obtained model parameters are saved, thus completing the transfer training and fine-tuning of the heating network balancing scheduling model.

6. The method for measuring and controlling room temperature of heat users according to claim 1, characterized in that, When performing transfer training on the heat network balancing scheduling model, a domain adaptation module is set in the heat network balancing scheduling model. The domain adaptation module uses three transfer learning algorithms, namely TCA, JDA and GFK, to map from source domain data to target domain data, thereby increasing the similarity between source domain data and target domain data.

7. The method for measuring and controlling room temperature of heat users according to claim 1, characterized in that, The convolutional neural network also includes a feature extraction module, which is used to extract features from the new source domain and target domain data after domain adaptation; The feature extraction module comprises three convolutional networks, each of which includes a convolutional layer, a pooling layer, and a ReLU activation layer. The first layer of the convolutional network has 32 convolutional kernels with a size of [3,3,1,32] and a stride of 1. The pooling layer uses 2*2 max pooling. After the input data passes through the first layer of the convolutional network, the size becomes [16*16]. The second convolutional network has 64 kernels with a size of [3,3,32,64] and a stride of 1. The pooling layer uses 2*2 max pooling. The output size after passing through the second convolutional network is [8*8]. The third convolutional network has 128 kernels of size [3,3,64,128] and a stride of 1. The pooling layer uses 2*2 max pooling. The output size after the second convolutional network is [4*4]. Then there is a fully connected layer of size [4*4*128,512] for feature integration.

8. The method for measuring and controlling room temperature of heat users according to claim 1, characterized in that, The convolutional neural network includes an input layer, a convolutional layer, an activation layer, a pooling layer, a fully connected layer, and a final output layer.

9. The method for measuring and controlling room temperature of heat users according to claim 1, characterized in that, The feedforward regulation of valves and pumps is performed using the aforementioned heat network balance optimization scheduling model, including: Since the lag time of the heating system is approximately equal to the fluid flow time in the pipe, the fluid flow time is selected as the feedforward adjustment time of the heating system, and the valves and pumps of the heating system are adjusted in advance. The feedforward adjustment time is calculated as follows: Where L is the pipe length, v w Let D be the velocity of the fluid in the pipe. w G is the diameter of the pipe. w This represents the flow rate of the fluid in the pipe.

Images