A Deep Learning-Based Method and Device for Optimizing and Controlling Heating Supply

Abstract

This invention discloses a heating optimization and control method and device based on deep learning. It acquires the time characteristics, outdoor temperature, and secondary network water supply temperature set for the time to be calculated, and constructs a first dataset. Using each set of data in the first dataset as input information, a first neural network model determines the user's predicted indoor temperature. A second dataset is constructed based on the predicted indoor temperature, outdoor temperature, and secondary network water supply temperature range. Using each set of data in the second dataset as input information, a second neural network model determines the corresponding predicted secondary network return water temperature set. The final secondary network water supply temperature is calculated based on the secondary network water supply temperature set and the predicted secondary network return water temperature set. This invention can improve the calculation accuracy of the secondary network water supply temperature while ensuring the indoor temperature, thus avoiding energy waste.

Description

Technical Field

[0001] This invention belongs to the field of centralized heating system water supply temperature calculation technology, and particularly relates to a heating optimization and control method and device based on deep learning. Background Technology

[0002] Centralized heating is an important infrastructure for building modern cities. It not only provides cities with a stable, reliable, and high-quality heat source, improving living conditions, but also saves energy, reduces energy consumption, reduces urban pollution, and protects the ecological environment.

[0003] In centralized heating systems, due to the complexity of users' heat demand, and the influence of multiple factors such as the structure of heat exchange stations and secondary pipe networks, the structure of heating users' buildings, insulation conditions, users' heating habits, and outdoor weather, traditional heating optimization and control methods are mostly determined by the experience of dispatchers or some simple calculation formulas.

[0004] However, the adjustment results estimated by the above methods are often difficult to achieve the desired accuracy, which can easily lead to users having room temperature that is not up to standard or that is too high. When the room temperature is too high, it will lead to energy waste on the one hand, and the excessive temperature will also reduce user comfort on the other hand. Summary of the Invention

[0005] The purpose of this invention is to provide a heating optimization and control method and device based on deep learning to improve the control accuracy of the heating system.

[0006] This invention adopts the following technical solution: a heating optimization and control method based on deep learning, comprising the following steps:

[0007] Obtain the time features, outdoor temperature, and secondary network water supply temperature set for the time to be calculated, and construct the first dataset; wherein, the time features include the monthly, weekly, and time series of the time to be calculated;

[0008] Using each set of data in the first dataset as input information, the predicted indoor temperature for the user is determined through the first neural network model.

[0009] A second dataset was constructed based on the predicted indoor temperature, outdoor temperature, and secondary network water supply temperature range.

[0010] Using each set of data in the second dataset as input information, the corresponding predicted secondary network return water temperature set is determined through the second neural network model.

[0011] The final secondary network supply temperature is calculated based on the set of secondary network supply water temperatures and the predicted set of secondary network return water temperatures.

[0012] Furthermore, constructing the first dataset includes:

[0013] Each pre-selected value of the secondary network water supply temperature in the secondary network water supply temperature set is combined with time characteristics and outdoor temperature as a set of data in the first dataset.

[0014] Furthermore, the first neural network model is a trained indoor temperature prediction model, and the training method for the indoor temperature prediction model is as follows:

[0015] Construct the first training dataset. Each set of data in the first training dataset includes the secondary network water supply temperature, outdoor temperature, time characteristics, and actual indoor temperature.

[0016] Input each set of data from the first training dataset into the indoor temperature prediction model to obtain the predicted indoor temperature output by the indoor temperature prediction model.

[0017] The indoor temperature prediction model is optimized based on the actual indoor temperature and the predicted indoor temperature until the indoor temperature prediction model converges.

[0018] Furthermore, determining the user's predicted indoor temperature using the first neural network model includes:

[0019] For each set of data in the first dataset, the first neural network model outputs the corresponding predicted indoor temperature;

[0020] Select the minimum predicted indoor temperature from several predicted indoor temperatures that satisfies the target indoor temperature as the predicted indoor temperature.

[0021] Furthermore, a second dataset was constructed based on the predicted indoor temperature, outdoor temperature, and secondary network water supply temperature, including:

[0022] Based on the predicted indoor temperature, a pre-selected value for the secondary network water supply temperature is chosen from the secondary network water supply temperature set.

[0023] The pre-selected value of the secondary network water supply temperature is combined with the predicted indoor and outdoor temperatures to form a set of data in the second dataset.

[0024] Furthermore, the second neural network model is a trained secondary network return water temperature prediction model. The training method for the secondary network return water temperature prediction model is as follows:

[0025] Construct a second training dataset. Each set of data in the second training dataset includes the secondary network water supply temperature, the predicted indoor temperature, the outdoor temperature, and the actual secondary network return water temperature.

[0026] Input each set of data in the second training dataset into the secondary network return water temperature prediction model to obtain the predicted secondary network return water temperature output by the secondary network return water temperature prediction model.

[0027] The secondary network return water temperature prediction model is optimized based on the predicted and actual secondary network return water temperatures until the model converges.

[0028] Furthermore, the final secondary network supply temperature is calculated based on the secondary network supply temperature set and the predicted secondary network return temperature set, including:

[0029] For each pre-selected value of the secondary network supply water temperature, select the corresponding predicted secondary network return water temperature from the predicted secondary network return water temperature set.

[0030] Calculate the temperature difference between the pre-selected secondary network supply water temperature and the corresponding predicted secondary network return water temperature.

[0031] Furthermore, after calculating the temperature difference between the pre-selected secondary network supply water temperature and the corresponding predicted secondary network return water temperature, the process also includes:

[0032] Select the pre-selected secondary network water supply temperature corresponding to the minimum temperature difference as the final secondary network water supply temperature.

[0033] Furthermore, obtaining the secondary network water supply temperature data includes:

[0034] Obtain the temperature range of the secondary water supply network;

[0035] A set of secondary network water supply temperatures is generated based on a predetermined step size and the secondary network water supply temperature range.

[0036] Another technical solution of the present invention: a heating optimization and control device based on deep learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above methods.

[0037] The beneficial effects of this invention are as follows: This invention predicts the indoor temperature through a first neural network model, then predicts the secondary network return water temperature through a second neural network model, and finally accurately calculates the secondary network supply water temperature by combining the predicted secondary network return water temperature and the corresponding secondary network supply water temperature. This can improve the calculation accuracy of the secondary network supply water temperature while ensuring the indoor temperature, thereby improving the control accuracy of the heating system and avoiding energy waste. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of data comparison during the training process of the first neural network model in an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of data comparison during the training process of the second neural network model in an embodiment of the present invention. Detailed Implementation

[0040] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0041] Unlike unattended or automated control systems, smart heating is characterized by intelligent analysis, self-learning, and predictive control. Its goal is to achieve self-awareness of the heating system's state, establish logical relationships between heat sources, the heating network, and heat users, and deliver the required heat energy to the correct location at the appropriate time. Therefore, it is necessary to propose a big data learning method for smart heating systems, focusing on energy consumption.

[0042] The goal of heating regulation is to meet people's thermal comfort needs, provide heating on demand, and reduce the energy consumption of the heating system. Therefore, the core task of heating regulation is to control room temperature based on the thermal comfort needs of heat users. This means achieving the goal of smart heating through real-time analysis, scientific decision-making, and precise execution of data from real-time room temperature monitoring.

[0043] Existing big data-driven heating control methods often deviate from users' target indoor temperature values, and the predicted heating system energy consumption cannot guarantee a minimum. Therefore, this invention proposes a big data algorithm model that, while ensuring room temperature, aims to minimize heating system energy consumption. This model can both satisfy users' room temperature requirements and obtain the operating parameters for minimizing heating system energy consumption, achieving the goal of minimizing energy consumption by adjusting supply and return water temperatures based on room temperature monitoring data.

[0044] This invention discloses a heating optimization and control method based on deep learning, comprising the following steps: acquiring the time features, outdoor temperature, and secondary network water supply temperature set of the time to be calculated, and constructing a first dataset; wherein, the time features include the monthly, weekly, and hourly sequence of the time to be calculated; using each set of data in the first dataset as input information, determining the user's predicted indoor temperature through a first neural network model; constructing a second dataset based on the predicted indoor temperature, outdoor temperature, and secondary network water supply temperature range; using each set of data in the second dataset as input information, determining the corresponding predicted secondary network return water temperature set through a second neural network model; and calculating the final secondary network water supply temperature based on the secondary network water supply temperature set and the predicted secondary network return water temperature set.

[0045] This invention uses a first neural network model to predict the indoor temperature, a second neural network model to predict the secondary network return water temperature, and finally combines the predicted secondary network return water temperature with the corresponding secondary network supply water temperature to accurately calculate the secondary network supply water temperature. This can improve the calculation accuracy of the secondary network supply water temperature while ensuring the indoor temperature and avoid energy waste.

[0046] As a specific implementation, obtaining the secondary network water supply temperature set includes: obtaining the secondary network water supply temperature range; and generating the secondary network water supply temperature set based on a predetermined step size and the secondary network water supply temperature range. Specifically, the secondary network water supply temperature range is typically 35–45℃. In this embodiment, the predetermined step size is set to 0.5, i.e., the secondary network water supply temperature set is {35.0℃, 35.5℃, 36.0℃, 36.5℃, 37.0℃, 37.5℃, 38.0℃, 38.5℃, 39.0℃, 39.5℃, 40.0℃, 40.5℃, 41.0℃, 41.5℃, 42.0℃, 42.5℃, 43.0℃, 43.5℃, 44.0℃, 44.5℃, 45.0℃}. Of course, different step sizes can be selected according to specific needs to obtain different secondary network water supply temperature sets.

[0047] In one embodiment, constructing the first dataset includes: combining each pre-selected value of the secondary network water supply temperature in the secondary network water supply temperature set with time characteristics and outdoor temperature as a set of data in the first dataset.

[0048] As a specific implementation, the first neural network model is a trained indoor temperature prediction model. The training method of the indoor temperature prediction model is as follows: construct a first training dataset, in which each set of data includes the secondary network water supply temperature, outdoor temperature, time characteristics, and actual indoor temperature; input each set of data from the first training dataset into the indoor temperature prediction model to obtain the predicted indoor temperature output by the indoor temperature prediction model; optimize the indoor temperature prediction model based on the actual indoor temperature and the predicted indoor temperature until the indoor temperature prediction model converges.

[0049] The first neural network model in this embodiment of the invention can be a deep learning network model. Deep learning (DL) is a new research direction in the field of machine learning (ML). It has been introduced into machine learning to bring it closer to its original goal—artificial intelligence (AI). Deep learning learns the inherent patterns and representational layers of sample data, and its common applications include text recognition, image recognition, and sound recognition.

[0050] Deep learning is a complex machine learning algorithm that has achieved significant results in search technology, data mining, machine translation, natural language processing, multimedia learning, speech recognition, recommendation and personalization technologies, and other related fields. Deep learning enables machines to mimic human activities such as sight, hearing, and thought, solving many complex pattern recognition problems and greatly advancing artificial intelligence technologies. A key idea of ​​deep learning is to achieve end-to-end modeling, also known as black-box modeling. Through self-learning techniques and upgrades in hardware computing power, deep learning can find statistical patterns in massive amounts of data and achieve astonishing accuracy in predictions. Therefore, it has enormous potential for engineering applications.

[0051] In this embodiment of the invention, the deep learning neural network method includes various types of neural networks, such as BP neural networks, RNN neural networks, RBF neural networks, LSTM neural networks, etc. Neural network methods can achieve excellent predictive modeling results, thereby completing the establishment of a predictive model. Different neural network methods have their own advantages and limitations. Therefore, in practical implementation, technicians can generally choose a specific neural network method to implement predictive model building as needed.

[0052] Before modeling after data collection, the data needs to be properly processed, including data quality inspection, data analysis, data visualization, data EDA, data preprocessing, and dataset construction.

[0053] In one embodiment, determining the user's predicted indoor temperature using a first neural network model includes: for each set of data in a first dataset, the first neural network model outputs the corresponding predicted indoor temperature; and selecting the minimum value among several predicted indoor temperatures that satisfies the target indoor temperature as the predicted indoor temperature.

[0054] Specifically, constructing a second dataset based on predicted indoor temperature, outdoor temperature, and secondary network water supply temperature includes: selecting a pre-selected value for the secondary network water supply temperature from the set of predicted indoor temperature; and combining the pre-selected value of the secondary network water supply temperature with the predicted indoor temperature and outdoor temperature as a set of data in the second dataset.

[0055] In addition, the second neural network model is a trained secondary network return water temperature prediction model. The training method for the secondary network return water temperature prediction model is as follows: a second training dataset is constructed, and each set of data in the second training dataset includes the secondary network supply water temperature, the predicted indoor temperature, the outdoor temperature, and the actual secondary network return water temperature; each set of data in the second training dataset is input into the secondary network return water temperature prediction model to obtain the predicted secondary network return water temperature output by the secondary network return water temperature prediction model; the secondary network return water temperature prediction model is optimized based on the predicted secondary network return water temperature and the actual secondary network return water temperature until the secondary network return water temperature prediction model converges.

[0056] In one embodiment, calculating the final secondary network supply temperature based on the secondary network supply temperature set and the predicted secondary network return temperature set includes: for each pre-selected value of the secondary network supply temperature, selecting the corresponding predicted secondary network return temperature from the predicted secondary network return temperature set; and calculating the temperature difference between the pre-selected value of the secondary network supply temperature and the corresponding predicted secondary network return temperature.

[0057] In addition, after calculating the temperature difference between the secondary network supply water temperature and the corresponding predicted secondary network return water temperature, the process also includes: selecting the pre-selected value of the secondary network supply water temperature corresponding to the minimum temperature difference as the final secondary network supply water temperature.

[0058] After obtaining the final secondary network water supply temperature, the control target, i.e., the secondary network water supply temperature, is sent to SCADA. SCADA sends the target to PLC, PLC sends the target to the equipment for execution, the equipment feeds back the execution results to PLC, PLC feeds back to SCADA, SCADA feeds back to the platform, and the platform presents the process and results to the user for viewing.

[0059] To facilitate understanding of this invention, a centralized heating project for residential buildings in Laiwu is used as an example to illustrate the implementation process of the technical solution of this invention in detail.

[0060] Sensors were installed at the project site to collect relevant data from the regional centralized heating system during the 2021-2022 heating season, with data collected every minute. This data included, but was not limited to, the following attribute points: secondary network supply water temperature, secondary network return water temperature, outdoor temperature, and indoor temperature. The collected data was stored in a time-series database on a local server and also categorized and stored in a cloud-based energy data warehouse for later analysis and modeling.

[0061] In this embodiment, the BP neural network algorithm is selected as the framework for the first and second neural network models based on the characteristics of residential buildings and the sample size. The network structure is then adaptively adjusted according to the dimensions of the input and output information.

[0062] The BP neural network algorithm has the following significant advantages:

[0063] 1) Nonlinear mapping capability: Backpropagation (BP) neural networks essentially implement a mapping function from input to output. Mathematical theory proves that a three-layer neural network can approximate any nonlinear continuous function with arbitrary precision. This makes it particularly suitable for solving problems with complex internal mechanisms; that is, BP neural networks have strong nonlinear mapping capabilities.

[0064] 2) Self-learning and adaptive capabilities: During training, BP neural networks can automatically extract the "reasonable rules" between outputs and output data through learning, and adaptively memorize the learned content into the network weights. That is, BP neural networks have a high degree of self-learning and adaptive capabilities.

[0065] 3) Generalization ability: Generalization ability refers to the ability of a learning model to correctly learn the learned objects, and also to correctly predict unseen patterns or noisy objects after training. In other words, a BP neural network has the ability to apply its learning results to new knowledge.

[0066] 4) Fault tolerance: When a BP neural network suffers damage to a local or partial number of neurons, it does not significantly affect the overall training results. In other words, the system can still function normally even when subjected to local damage. Therefore, BP neural networks have good fault tolerance.

[0067] Next, the collected data will be read from the database, specifically the indoor temperature, outdoor temperature, secondary network water supply temperature, and secondary network return water temperature from 00:00:00 on December 1, 2021 to 00:00:00 on February 27, 2022.

[0068] In this embodiment, the indoor temperature is obtained by installing indoor temperature sensors to collect actual temperature data. A total of 116 households were sampled, and the indoor temperature at that moment was calculated by averaging the indoor temperature of each household at the same time.

[0069] Then, the completeness of the collected data was checked. A long-term series was collected at 1-minute intervals from the initial time of 00:00:00 on December 1, 2021 to 00:00:00 on February 27, 2022, to examine the data at each time point. It was found that the data at 05:00:00 on December 31, 2021 was missing due to a communication interruption in the data acquisition equipment. Therefore, the indoor temperature, outdoor temperature, secondary network supply water temperature, and secondary network return water temperature at this time point were compensated using the mean. Outlier detection: The 3-sigma rule was used for detection, and outlier data were replaced with the mean. Additionally, during data preprocessing, a small number of secondary network supply water temperatures below 34℃ were found. To ensure the effectiveness of the training prediction model, these data samples were removed.

[0070] Dataset Preparation: Weekly and time series data were calculated based on time points. Weekly data ranges from 0 to 6, representing Sunday to Saturday respectively (e.g., 1 represents Monday, 2 represents Tuesday, and so on). Time series data ranges from 0 to 23, with data from 0:00 to 1:00 uniformly encoded as 0. The collected historical data was compiled into a dataset of 126,722 samples. The test and training sets were split according to a 2:8 ratio. The training set contains 101,378 samples, and the test set contains 25,344 samples. A sample dataset format is shown in Table 1 below, where Systime represents time, w represents weekly data, h represents time series, tn represents indoor temperature, tw represents outdoor temperature, tg represents secondary network supply water temperature, and th represents secondary network return water temperature.

[0071] Table 1

[0072] 2021 / 12 / 1 0:00 3 0 21.77561 1 44.1 38.1 2021 / 12 / 1 0:01 3 0 21.77561 1 43.9 38.1 2021 / 12 / 1 0:02 3 0 21.77561 -1 44.1 38.1 2021 / 12 / 1 0:03 3 0 21.77561 -1 44.2 38.1 2021 / 12 / 1 0:04 3 0 21.77561 -1 44.2 38.1 2021 / 12 / 1 0:05 3 0 21.77561 -1 44 38.1 2021 / 12 / 1 0:06 3 0 21.77561 -1 43.9 38.1

[0073] Next, the data is used to model the control system.

[0074] In heating control systems, meeting users' heat load demands is a key requirement. In heating systems, the way to judge whether heat demand is met is to see if the indoor temperature reaches the target set temperature. Generally, an indoor temperature above 18°C ​​is considered to meet heating demand. In this embodiment, to improve user thermal comfort, the target indoor temperature is set at 20°C.

[0075] The first neural network model is constructed as follows: This model is an indoor temperature prediction model, designed to describe the relationship between secondary network water supply temperature, outdoor temperature changes, and indoor temperature changes. In the heating system, indoor temperature is affected by both outdoor temperature and secondary network water supply temperature. Furthermore, indoor temperature changes are also influenced by user behavior, such as whether windows are opened for ventilation and whether the user is at home. User behavior is reflected at different times, such as differences in behavior on weekdays and weekends, and different heating demands in the morning and evening. Therefore, when inputting features, time-series features are extracted and listed separately as important features. In this embodiment, the time features include weekly sequence (w), time sequence (h), and monthly sequence (m).

[0076] A three-layer backpropagation (BP) neural network model with 300 neurons in each layer was constructed. The activation function was the ReLU function. The optimization algorithm was stochastic gradient descent with an initial learning rate of 0.00001 and a maximum number of iterations of 1000.

[0077] The input dataset includes historical secondary network water supply temperature, outdoor temperature, weekly sequence, time sequence, and monthly sequence. Through training, the output is used to predict indoor temperature, thereby constructing the first neural network model of the influence of secondary network water supply temperature, outdoor temperature, and user heating characteristics on indoor temperature.

[0078] After iterative optimization training, the first neural network model was evaluated using the MSE and R2 scores as evaluation metrics, resulting in an MSE score of 5.78 and an R2 score of 0.78.

[0079] The comparison chart of specific model prediction results is shown below. Figure 1 As shown in the figure, the comparison between the actual indoor temperature and the predicted indoor temperature of 37 groups can be seen.

[0080] A second neural network model is constructed: This model is a secondary network return water temperature prediction model, designed to establish the relationship between the secondary network supply water temperature, indoor temperature, outdoor temperature, and secondary network return water temperature. In a heating system, under a constant flow rate, the system generates the least heat, i.e., consumes the least energy, when the supply and return water temperature difference is minimized.

[0081] A three-layer backpropagation (BP) neural network model with 300 neurons in each layer was constructed. The activation function was the ReLU function. The optimization algorithm was stochastic gradient descent with an initial learning rate of 0.00001 and a maximum number of iterations of 1000.

[0082] Input the secondary network supply water temperature, indoor temperature, and outdoor temperature, and output the secondary network return water temperature to construct a return water temperature prediction model. After iterative optimization and training, the model is evaluated using the MSE and R2 scores as evaluation indicators, and the MSE score is 2.77 and the R2 score is 0.98.

[0083] The comparison chart of specific model prediction results is shown below. Figure 2 As shown in the figure, the actual secondary network return water temperature and the predicted secondary network return water temperature of the 30 groups can be seen from the figure.

[0084] Then, the first neural network model is used to predict the impact of different secondary network water supply temperatures on indoor temperature. Further, 21 secondary network water supply temperature values ​​(35-45℃ at 0.5℃ intervals) are input into the trained first neural network model to predict the time characteristics at different time points, and the indoor temperatures corresponding to different secondary network water supply temperatures are predicted, as shown in Table 2. In this embodiment, the predicted indoor temperature is selected as 20.11954, approximately equal to 20.1℃.

[0085] Table 2

[0086]

[0087]

[0088] Then, the pre-selected value of the secondary network water supply temperature (i.e., the secondary network water supply temperature value is greater than or equal to the secondary network water supply temperature value corresponding to the predicted indoor temperature), the outdoor temperature, and the predicted indoor temperature are input into the second neural network model to obtain the predicted return water temperature. The prediction results are shown in Table 3.

[0089] Finally, the pre-selected value of the secondary network supply water temperature corresponding to the minimum supply and return water temperature difference is found as the target control value (the smaller the temperature difference, the lower the energy consumption of the heating system), which is 40.5℃.

[0090] Table 3

[0091]

[0092]

[0093] It should be noted that although in this embodiment, the supply and return water temperature difference is linearly related to the pre-selected value of the secondary network supply water temperature, this is only a relatively special embodiment. In reality, the two are not linearly related. For example, sometimes the supply and return water temperature difference is at its minimum when the pre-selected value of the secondary network supply water temperature is at its maximum.

[0094] Therefore, the secondary network water supply temperature was selected as 40.5℃. The platform sends the control target, i.e., the water supply temperature, to the SCADA system, which then sends it to the PLC. The PLC sends it to the equipment for execution, the equipment feeds back the execution results to the PLC, the PLC feeds back to the SCADA system, the SCADA system feeds back to the platform, and the platform presents the process and results to the user.

[0095] The above-mentioned project has been in operation for two heating seasons. The big data learning method was not used for optimization and control in the 2020-2021 heating season. Data from 3.1 to 3.8 of a certain heat exchange station in the 2020-2021 and 2021-2022 heating seasons were selected for energy consumption comparison. The specific data is shown in Table 4 below.

[0096] Table 4

[0097]

[0098]

[0099] The table above shows that the cumulative heat consumption from March 1st to March 7th during the 2020-2021 heating season was 197.26 GJ, with an average outdoor temperature of 4.6℃; and the cumulative heat consumption from March 1st to March 7th during the 2021-2022 heating season was 142.01 GJ, with an average outdoor temperature of 8.7℃.

[0100] Based on the actual heat consumption data of the 2020-2021 heating season, and assuming an indoor temperature of 20℃, the theoretical heat consumption under the condition of an average outdoor temperature of 8.7℃ in the 2021-2022 heating season is (20-8.7)*197.26*15142 (area) / (20-4.6)*14043 (last year's area) = 156.07 GJ. To increase the comparability of the two, the cumulative heat consumption from March 1st to March 7th of the 2020-2021 heating season is used. The heat consumption of 197.26 GJ is multiplied by a conversion factor, namely (20-8.7)*15142 / (20-4.6)*14043, where (20-8.7) is the temperature rise range for the 2020-2021 heating season, 15142 is the heating area for the 2020-2021 heating season, (20-4.6) is the temperature rise range for the 2021-2022 heating season, and 14043 is the heating area for the 2021-2022 heating season. By increasing the above conversion factor, the difference in external conditions between different heating seasons can be minimized as much as possible, thereby ensuring that the energy consumption of the two heating seasons can be compared. Finally, the relative energy saving rate of the heating season using the method of this invention and the heating season without the method of this invention is (156.07-142.01) / 156.07 = 9%, that is, the energy consumption is reduced by 9%. Based on the heating company's selling price of 46.84 yuan / GJ, a total of 658.57 yuan was saved over these 7 days. Calculated by heating area, a typical residential building can save 100 yuan in heating costs per day. If this method is applied to the entire city, taking a city with a population of 10 million as an example, the total cost savings would be immeasurable, approximately 3 to 4 million yuan per day.

[0101] The present invention also discloses a heating optimization and control device based on deep learning, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method.

[0102] The present invention implements all or part of the processes in the methods of the above embodiments, which can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a storage device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0103] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0104] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

Claims

1. A heating optimization and control method based on deep learning, characterized in that, Includes the following steps: Obtain the time features, outdoor temperature, and secondary network water supply temperature set for the time to be calculated, and construct the first dataset; wherein, the time features include the monthly, weekly, and temporal sequence of the time to be calculated; Using each set of data in the first dataset as input information, the predicted indoor temperature for the user is determined through the first neural network model. A second dataset is constructed based on the predicted indoor temperature, outdoor temperature, and secondary network water supply temperature range. Using each set of data in the second dataset as input information, the corresponding predicted secondary network return water temperature set is determined through the second neural network model; The final secondary network water supply temperature is calculated based on the set of secondary network supply water temperatures and the predicted set of secondary network return water temperatures. The predicted indoor temperature for users is determined using the first neural network model, including: For each set of data in the first dataset, the first neural network model outputs the corresponding predicted indoor temperature. The minimum value that satisfies the target indoor temperature is selected from several predicted indoor temperatures as the predicted indoor temperature; The final secondary network supply temperature is calculated based on the set of secondary network supply water temperatures and the predicted set of secondary network return water temperatures, including: For each of the pre-selected values ​​of the secondary network water supply temperature, select the corresponding predicted secondary network return water temperature from the set of predicted secondary network return water temperatures. Calculate the temperature difference between the pre-selected secondary network water supply temperature and the corresponding predicted secondary network return water temperature; The pre-selected value of the secondary network water supply temperature corresponding to the minimum temperature difference is selected as the final secondary network water supply temperature.

2. The heating optimization and control method based on deep learning as described in claim 1, characterized in that, Building the first dataset includes: Each pre-selected value of the secondary network water supply temperature in the secondary network water supply temperature set is combined with the time feature and the outdoor temperature to form a set of data in the first dataset.

3. A heating optimization and control method based on deep learning as described in claim 1 or 2, characterized in that, The first neural network model is a trained indoor temperature prediction model, and the training method for the indoor temperature prediction model is as follows: Construct a first training dataset, in which each set of data includes secondary network water supply temperature, outdoor temperature, time characteristics, and actual indoor temperature; Input each set of data in the first training dataset into the indoor temperature prediction model to obtain the predicted indoor temperature output by the indoor temperature prediction model. The indoor temperature prediction model is optimized based on the actual indoor temperature and the predicted indoor temperature until the indoor temperature prediction model converges.

4. The heating optimization and control method based on deep learning as described in claim 3, characterized in that, The second dataset, constructed based on the predicted indoor temperature, outdoor temperature, and secondary network water supply temperature sets, includes: Based on the predicted indoor temperature, a pre-selected value for the secondary network water supply temperature is selected from the set of secondary network water supply temperatures. The pre-selected value of the secondary network water supply temperature is combined with the predicted indoor and outdoor temperatures to form a set of data in the second dataset.

5. The heating optimization and control method based on deep learning as described in claim 4, characterized in that, The second neural network model is a trained secondary network return water temperature prediction model. The training method for the secondary network return water temperature prediction model is as follows: Construct a second training dataset, in which each set of data includes the secondary network water supply temperature, the predicted indoor temperature, the outdoor temperature, and the actual secondary network return water temperature; Input each set of data in the second training dataset into the secondary network return water temperature prediction model to obtain the predicted secondary network return water temperature output by the secondary network return water temperature prediction model. The secondary network return water temperature prediction model is optimized based on the predicted secondary network return water temperature and the actual secondary network return water temperature until the secondary network return water temperature prediction model converges.

6. The heating optimization and control method based on deep learning as described in claim 5, characterized in that, Obtaining the secondary network water supply temperature set includes: Obtain the temperature range of the secondary water supply network; The secondary network water supply temperature set is generated based on a predetermined step size and the secondary network water supply temperature range.

7. A heating optimization and control device based on deep learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 6.

Images