Hot water boiler intelligent control system and method based on thermal load prediction

Abstract

The invention provides a hot water boiler intelligent control system and method based on thermal load prediction, and relates to the field of intelligent control. Boiler operation state data collected by a sensor matrix is used as a main variable and external environment information is used as a covariable by using a data analysis and processing mode based on artificial intelligence; respectively extracting thermal load main variable time sequence coding characteristics and thermal load covariable time sequence coding characteristics from the main variable and the covariable; based on dynamic memory interaction representation between thermal load main variable time sequence coding characteristics and thermal load covariable time sequence coding characteristics, thermal load demand intelligent estimation is realized, and boiler fuel supply quantity is adjusted in real time according to the thermal load demand intelligent estimation. In this way, the change of the external environment can be responded in real time, and the efficient operation state can be kept in the complex and changeable environment.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to an intelligent control system and method for a hot water boiler based on heat load prediction. Background Art

[0002] Hot water boilers are widely used in industrial, commercial and residential areas to provide hot water or steam to meet the needs of heating, process heating, etc. With the rising energy costs and increasingly stringent requirements for environmental protection, improving the energy efficiency of hot water boiler systems, reducing fuel consumption and lowering emissions have become key goals for the development of the industry.

[0003] Traditional hot water boiler control systems rely on fixed set points and simple feedback mechanisms to regulate fuel supply, which is inadequate when faced with complex operating environments and fluctuating heat load demands. Due to the lack of sophisticated analysis of real-time data, it is difficult for the system to optimize fuel use, resulting in energy waste and unnecessary emissions. At the same time, changes in external environmental conditions such as temperature, humidity, and wind speed can also significantly affect heat load demand, but traditional systems often do not have real-time response capabilities and cannot dynamically adjust to these changes, further exacerbating the problem of inefficiency. In addition, the lag and limited adaptability of traditional systems limit their performance under variable conditions, causing the boiler to operate in a non-optimal state, increasing maintenance costs and the risk of failure.

[0004] Therefore, an intelligent control scheme for hot water boilers based on heat load prediction is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a hot water boiler intelligent control system and method based on heat load prediction.

[0006] According to one aspect of the present application, a hot water boiler intelligent control system based on heat load prediction is provided, which includes: a boiler operation status data acquisition module, which is used to collect boiler operation status data using a sensor matrix; an external environment information acquisition module, which is used to collect external environment information; a data time series encoding module, which is used to take the boiler operation status data as the main variable and the external environment information as the covariate, and extract the heat load main variable time series encoding characteristics and the heat load covariate time series encoding characteristics from the main variable and the covariate; a heat load main-covariate dynamic time series interaction module, which is used to perform dynamic memory feature attention fusion on the heat load main variable time series encoding characteristics and the heat load covariate time series encoding characteristics to obtain the heat load main-covariate dynamic memory Time series interaction fusion representation, wherein the heat load main-covariate dynamic time series interaction module includes: a time series implicit collaborative feature analysis unit, used to perform collaborative implicit feature extraction on the heat load main variable time series coding features and the heat load covariate time series coding features to obtain the heat load main-covariate time series implicit collaborative features; a feature modulation fusion unit, used to perform attention modulation fusion on the heat load main variable time series coding features and the heat load covariate time series coding features according to the heat load main-covariate time series implicit collaborative features to obtain the heat load main-covariate dynamic memory time series interaction fusion representation; a boiler fuel supply adjustment module, used to adjust the boiler fuel supply according to the heat load main-covariate dynamic memory time series interaction fusion representation.

[0007] According to another aspect of the present application, a hot water boiler intelligent control method based on heat load prediction is provided, which includes: using a sensor matrix to collect boiler operation status data; collecting external environmental information; taking the boiler operation status data as the main variable and the external environmental information as the covariate, extracting the heat load main variable time series coding features and the heat load covariate time series coding features from the main variable and the covariate; performing dynamic memory feature inter-attention fusion on the heat load main variable time series coding features and the heat load covariate time series coding features to obtain a heat load main-covariate dynamic memory time series interactive fusion representation, including: performing collaborative implicit feature extraction on the heat load main variable time series coding features and the heat load covariate time series coding features to obtain a heat load main-covariate time series implicit collaborative feature; performing attention modulation fusion on the heat load main variable time series coding features and the heat load covariate time series coding features according to the heat load main-covariate time series implicit collaborative feature to obtain the heat load main-covariate dynamic memory time series interactive fusion representation; adjusting the boiler fuel supply according to the heat load main-covariate dynamic memory time series interactive fusion representation.

[0008] Compared with the prior art, the hot water boiler intelligent control system and method based on heat load prediction provided by the present application uses artificial intelligence-based data analysis and processing methods to use the boiler operation status data collected by the sensor matrix as the main variable and the external environment information as the covariate, and then extracts the heat load main variable time series coding characteristics and the heat load covariate time series coding characteristics from the main variable and the covariate respectively, so as to realize intelligent estimation of heat load demand based on the dynamic memory interaction between the heat load main variable time series coding characteristics and the heat load covariate time series coding characteristics and adjust the boiler fuel supply in real time accordingly. In this way, it is possible to respond to changes in the external environment in real time, ensure that the boiler is always operating in the optimal state, help reduce energy waste and unnecessary emissions, and maintain an efficient operating state in a complex and changing environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0010] In the attached picture: Figure 1 4 is a block diagram of a hot water boiler intelligent control system based on heat load prediction according to an embodiment of the present application.

[0011] Figure 2 The block diagram is a heat load main-covariate dynamic time series interaction module in a hot water boiler intelligent control system based on heat load prediction according to an embodiment of the present application.

[0012] Figure 3 It is a block diagram of a feature modulation fusion unit in a hot water boiler intelligent control system based on heat load prediction according to an embodiment of the present application.

[0013] Figure 4 The block diagram is a boiler fuel supply adjustment module in a hot water boiler intelligent control system based on heat load prediction according to an embodiment of the present application.

[0014] Figure 5 The present invention is a flowchart of a hot water boiler intelligent control method based on heat load prediction according to an embodiment of the present application. DETAILED DESCRIPTION

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0016] Hot water boilers play an important role in providing hot water and steam to meet heating and process heating needs in many fields, including industrial, commercial and residential areas. However, with the continuous rise in energy prices and increasingly stringent environmental protection standards, improving the energy efficiency of hot water boiler systems, reducing fuel consumption and emissions have become the core goals that the industry needs to achieve.

[0017] In the past, the control system of hot water boilers mainly relied on fixed set points and relatively simple feedback mechanisms to regulate fuel supply. However, this control strategy is not enough when faced with complex and changing operating environments and fluctuations in heat load demand. Due to the lack of in-depth analysis and processing of real-time data, it is difficult for the system to optimize fuel use, which leads to energy waste and unnecessary increase in emissions. At the same time, changes in external environmental factors such as temperature, humidity, and wind speed can also have a significant impact on heat load demand. Unfortunately, traditional control systems often lack the ability to respond in real time and cannot dynamically adjust according to these changes, further exacerbating the problem of low energy efficiency. In addition, the hysteresis and limited adaptive ability of traditional systems severely limit their performance under variable conditions, causing the boiler to operate in a non-optimal state for a long time, which undoubtedly increases the cost of maintenance and the risk of failure.

[0018] In response to the above technical problems, the technical concept of the present application is to use a sensor matrix to collect boiler operation status data and external environment information, and use artificial intelligence-based data analysis and processing methods to use the boiler operation status data as the main variable and the external environment information as the covariate, and then extract the heat load main variable time series coding characteristics and the heat load covariate time series coding characteristics from the main variable and the covariate respectively, so as to realize intelligent estimation of heat load demand and real-time adjustment of boiler fuel supply based on the dynamic memory interaction between the heat load main variable time series coding characteristics and the heat load covariate time series coding characteristics. In this way, it is possible to respond to changes in the external environment in real time, quickly adjust the fuel supply, ensure that the boiler is always operating in the optimal state, reduce energy waste and unnecessary emissions, and enable it to maintain an efficient operating state in a complex and changing environment.

[0019] Figure 1 FIG. 1 is a block diagram of an intelligent control system for hot water boilers based on heat load prediction according to an embodiment of the present application. Figure 1As shown, in the hot water boiler intelligent control system 100 based on heat load prediction, it includes: a boiler operation status data acquisition module 110, which is used to collect boiler operation status data using a sensor matrix; an external environment information acquisition module 120, which is used to collect external environment information; a data time series encoding module 130, which is used to take the boiler operation status data as the main variable and the external environment information as the covariate, and extract the heat load main variable time series encoding characteristics and the heat load covariate time series encoding characteristics from the main variable and the covariate; a heat load main-covariate dynamic time series interaction module 140, which is used to perform dynamic memory feature attention fusion on the heat load main variable time series encoding characteristics and the heat load covariate time series encoding characteristics to obtain a heat load main-covariate dynamic memory time series interaction fusion representation; a boiler fuel supply adjustment module 150, which is used to adjust the boiler fuel supply according to the heat load main-covariate dynamic memory time series interaction fusion representation.

[0020] In the embodiment of the present application, the boiler operation status data acquisition module 110 is used to collect boiler operation status data using a sensor matrix. It should be understood that the boiler operation status data includes boiler internal water temperature data, pressure data, water flow data and water level data. There is an interdependent relationship between these specific operation status data and the heat load demand. Specifically, the change of heat load demand will directly lead to the change of boiler operation status data. For example, when the heat load increases, the water temperature will drop (when the fuel supply has not been adjusted), the pressure will increase (due to the expansion of water), the flow rate may be accelerated (in order to transport more heat), and the water level will drop (increase in water evaporation). Conversely, by monitoring the changes in the boiler operation status data, the changes in the heat load demand can be inferred. For example, when it is monitored that the water temperature continues to drop, the pressure continues to rise, the flow rate decreases, or the water level drops abnormally, it can be judged that the heat load demand has increased, and the fuel supply needs to be adjusted accordingly to meet the demand. That is, the boiler operating status data collected by the sensor matrix can provide real-time, comprehensive and integrated understanding of the boiler's working conditions. By analyzing these data using algorithms, the heat load demand value of the hot water boiler can be decoded and the fuel supply can be quickly adjusted accordingly to ensure that the hot water boiler operates in the optimal state.

[0021] The process of collecting boiler operating status data using a sensor matrix is ​​intended to obtain comprehensive and accurate boiler operating information through precise monitoring methods. The following is a detailed description of its implementation process: First, sensor selection and installation. The water temperature sensor needs to be selected according to the boiler operating temperature range and accuracy requirements. For example, the water temperature of a small hot water boiler is between 0-100°C, and a platinum thermal resistor (Pt100) sensor is often used, with an accuracy of ±0.15°C. It is installed near the outlet or return port and must be completely immersed in water to avoid turbulent water flow or bubble areas. It must be fixed by welding or threaded connection to ensure good heat conduction. The pressure sensor should be selected according to the working pressure range. For a boiler of 0-1.6MPa, a diffused silicon pressure sensor with an accuracy of ±0.25%FS is more suitable. It is installed in the steam or water space, and the pressure port is perpendicular to the pipeline and unobstructed. It is connected with a flange or thread and equipped with a stop valve and a drain valve. The water flow sensor is selected according to the water flow range and pipe size. For systems with large flow and good water quality, an electromagnetic flow meter can be selected. It should be installed in the water inlet or outlet pipe. There should be enough straight pipe sections in front and behind. The front straight pipe section should not be less than 10 times the pipe diameter, and the rear straight pipe section should not be less than 5 times the pipe diameter. It should be installed horizontally. Common types of water level sensors include float type, static pressure type and ultrasonic type. For large hot water boilers, static pressure type can be selected. The accuracy is ±1cm and it should be installed near the water level gauge. For static pressure type, pay attention to the connection of the pressure port. For ultrasonic type, avoid obstacles and keep the transmitting and receiving surfaces clean.

[0022] Next is data acquisition and transmission. To build a data acquisition system, connect each sensor to the data acquisition module through a signal cable. The module must have multi-channel input, high-precision conversion and high-speed sampling rate, such as a 16-channel, 16-bit precision, 100kHz sampling rate module, to achieve simultaneous acquisition of multiple sensor data. The transmission method can be selected according to the actual situation. If the distance is short and the interference is small, RS-485 bus wired transmission can be used, which is stable and low-cost; Ethernet or Wi-Fi, 4G / 5G and other wireless transmission can be used for large distribution areas or remote monitoring. The data acquisition frequency is set according to the speed of parameter changes and the real-time requirements of the control system. Parameters that change quickly, such as water temperature and pressure, are 1-10 times per second, and parameters that change slowly, such as water level and water flow, are 1-5 times per minute.

[0023] Then comes data preprocessing and calibration. The collected data may contain noise or outliers, and needs to be preprocessed, such as using median filtering to remove water temperature data noise. Sensors should be calibrated regularly. Temperature sensors are compared with standard thermometers, and pressure sensors are calibrated with standard pressure sources. The calibration cycle depends on the stability of the sensor and the use environment, generally once a month or once a quarter. The error is recorded during calibration and compensated in the control system.

[0024] In terms of data storage and management, the storage media can be hard disk, solid state disk or cloud storage. Hard disk has large capacity and low cost, which is suitable for local storage of large amounts of historical data. SSD has fast read and write speed, which improves storage and reading efficiency. Cloud storage is convenient for remote backup and sharing, and is suitable for multi-site or distributed systems. At the same time, a data management strategy should be formulated to classify, archive and back up data. Classified storage is convenient for query and analysis. Expired data should be archived and deleted regularly to free up space. The data of the day should be backed up every day. Full backup and off-site storage or cloud storage should be performed every week to ensure data security and integrity.

[0025] Finally, there is system maintenance and troubleshooting. The appearance, connecting cables and measuring elements of sensors should be checked regularly. The protective sleeve of temperature sensors should be checked. The pressure tapping port of pressure sensors should be checked. Damaged parts should be cleaned or replaced in time. The data acquisition system should check the power supply, communication interface and hardware equipment. The software should be maintained to ensure compatibility and stability. When data acquisition is abnormal, the system log, cable connection and sensor signal should be checked for faults. In the communication system, the cables and connectors for wired transmission should be checked, and the signal strength and frequency band interference for wireless transmission should be checked. If necessary, the parameters should be adjusted or the equipment should be replaced to ensure stable and reliable transmission.

[0026] Through the above comprehensive and meticulous implementation process, boiler operation status data can be collected accurately and reliably, thus providing a solid data foundation for subsequent analysis.

[0027] In an embodiment of the present application, the external environment information acquisition module 120 is used to collect external environment information. It should be understood that the external environment information specifically refers to the temperature information, humidity information and wind speed information of the environment. Accordingly, considering that the heat load demand is not only determined by the internal state of the hot water boiler, the external environment also has a vital impact on it. Specifically, according to the basic principle of heat transfer, heat is always transferred from a high-temperature object to a low-temperature object. When the external environment temperature is low, the temperature difference between the inside and outside of the hot water boiler is large, and the heat conduction through the surface of the hot water boiler increases. At the same time, humidity and wind speed factors will enhance this heat transfer effect. The hot water boiler needs to provide enough heat to balance this heat loss, so the heat load demand of the hot water boiler is also closely related to the external environment information. By collecting external environment information, more factors affecting the heat load demand can be taken into consideration, which is conducive to a more comprehensive and accurate prediction of the heat load demand.

[0028] The implementation process of collecting external environmental information covers several key links, including the selection of environmental information collection equipment, data collection and transmission methods, data processing, data storage and management, and system maintenance and calibration. The following is a detailed description of its implementation process: First, the selection of environmental information collection equipment is the basis. In terms of air temperature sensors, common types include thermocouples, thermal resistors, and semiconductor temperature sensors. Thermocouples work based on the thermoelectric effect, thermal resistors use the characteristics of resistance value changing with temperature, and semiconductor temperature sensors have high sensitivity and fast response. In places where high accuracy is required and the environment is stable, Pt100 platinum resistance thermometers can be selected, and their measurement accuracy can reach ±0.15°C. When installing, it should be placed in a well-ventilated place without direct sunlight and can represent the ambient temperature, such as a cool place in a building or in a louvered box, at a height of 1.5-2 meters, and keep a distance from surrounding objects to prevent heat transfer from affecting accuracy.

[0029] Common humidity sensors are capacitive and resistive. Capacitive humidity sensors use the principle that dielectric capacitance changes with humidity. They have fast response, high accuracy, and good stability. They are often used in places with high humidity accuracy requirements, such as precision instrument workshops. The installation location should be well ventilated and avoid condensation. It is usually 1-1.5 meters away from the indoor wall. Pay attention to the direction to prevent airflow impact and contamination. Clean and calibrate regularly to ensure accuracy.

[0030] Common wind speed sensors include three-cup type and ultrasonic type. The three-cup type measures wind speed by rotating the wind cup, and has a simple and reliable structure; the ultrasonic type uses the principle that the propagation speed of ultrasonic waves is affected by wind speed, and has high accuracy and fast response, making it suitable for high-precision measurement occasions. The installation should be in an open and unobstructed place, away from buildings and trees, at a height of 10-20 meters, and ensure horizontal installation, and regularly check the mechanical parts and electrical connections.

[0031] Next is the data collection and transmission method. In terms of wired transmission, if the distance is short and the interference is small, RS-485 bus or Ethernet can be used. RS-485 has a long transmission distance and strong anti-interference, which is suitable for multi-sensor connection; Ethernet has a fast transmission speed and a large amount of data, which is convenient for remote monitoring. The sensor connects to the data acquisition device through the corresponding interface and then transmits it to the control system. When wiring, you should choose a suitable cable to ensure shielding, lay protection reasonably, and protect it with cable troughs and wire pipes. In terms of wireless transmission, Wi-Fi is suitable for short-distance high-speed environments, such as local areas in buildings; ZigBee low-power self-organizing networks are suitable for places where sensors are dispersed and power consumption requirements are low; 4G / 5G can be transmitted remotely at high speed for large-scale wide-area collection. Wireless sensor nodes must be equipped with communication modules and configured with parameters to ensure signal coverage. Signals can be enhanced by relaying or optimizing antennas. When using public networks, pay attention to traffic costs.

[0032] Then comes data processing and fusion. The collected data may contain noise, outliers or missing values, and need to be cleaned and verified. Cleaning can use filtering algorithms to remove noise, such as sliding average filtering to process wind speed data; outliers are identified and corrected by setting thresholds or comparing with historical data; missing values ​​are supplemented by interpolation algorithms. In terms of multi-source data fusion, temperature, humidity and wind speed are related, and weighted average method, Kalman filter method, neural network method, etc. can be used. For example, the weighted average method fuses temperature data and assigns weights according to the accuracy and reliability of the sensor; the neural network fuses the data of the three, learns the internal relationship and predicts the trend, and provides a basis for boiler control.

[0033] Data storage and management are also indispensable. The storage format can be CSV, JSON, etc., and the database management system can be selected according to the needs, such as MySQL, Oracle or SQLite. MySQL is stable, powerful and easy to use, suitable for large-scale storage; SQLite is lightweight and suitable for embedded or low-resource demand occasions. It is necessary to formulate a backup and recovery strategy, regularly perform full or incremental backups, store them locally, on external devices or in the cloud, and restore them when data is lost or damaged. At the same time, it is necessary to maintain and optimize the database, clean up useless data, optimize query statements and rebuild indexes.

[0034] Finally, there is system maintenance and calibration. Regularly maintain the acquisition equipment, check the sensor status, battery power (wireless sensor), communication module performance, etc., and clean the sensor surface. Check the cable connection of wired devices and the signal strength stability of wireless devices, and replace aging and damaged parts in time. The accuracy of the equipment may decrease during use, and regular calibration is required. Compare with standard equipment, such as calibrating the temperature sensor with a high-precision thermometer, etc., adjust the parameters according to the calibration results, and optimize the installation position and algorithm to improve accuracy and reliability, such as fine-tuning the sensor position according to the monitoring results.

[0035] Through the comprehensive implementation of the above, external environmental information can be effectively collected, providing strong support for the intelligent control of hot water boilers.

[0036] In an embodiment of the present application, the data time series encoding module 130 is used to take the boiler operating status data as the main variable and the external environment information as the covariate, and extract the heat load main variable time series encoding features and the heat load covariate time series encoding features from the main variable and the covariate. Specifically, the data time series encoding module includes: a data integration unit, which is used to integrate the boiler operating status data and the external environment information according to the time dimension to obtain a sequence of boiler operating status time series data and a sequence of external environment time series data; a data time series encoding feature generation unit, which is used to input the sequence of boiler operating status time series data and the sequence of external environment time series data into a time encoder based on a forward LSTM model to obtain a heat load main variable time series encoding feature vector as the heat load main variable time series encoding feature and a heat load covariate time series encoding feature vector as the heat load covariate time series encoding feature.

[0037] It should be understood that the boiler operating status data (such as temperature, pressure, flow, etc.) directly reflects the current working condition and heat load demand of the boiler. The changes in these parameters are the key factors that determine the performance and efficiency of the boiler, and are the "cause" that determines the generation of heat load, that is, their changes will lead to changes in heat load. In contrast, external environmental information (such as outdoor temperature, humidity, etc.) is usually uncontrollable or an indirect influencing factor, but they will affect the heat load demand and belong to the influencing factors before the "effect" in the causal chain. Therefore, in the technical solution of the present application, the boiler operating status data is used as the main variable and the external environmental information is used as the covariate. In this way, the boiler operating status data is used as the main variable, which can more directly reflect the dynamic behavior of the system, and the external environmental information is used as the covariate, which can help the model better capture those factors that are not directly controlled but are still important. Next, in order to better capture the change trend of the boiler operating status data (main variable) and the external environmental information (covariate) over time, such as rising, falling or periodic fluctuations, further, the heat load main variable time series encoding feature vector and the heat load covariate time series encoding feature vector are extracted from the main variable and the covariate. In particular, in a specific embodiment of the present application, first, the boiler operation state data and the external environment information are integrated according to the time dimension to arrange them according to the time dimension, and a sequence of boiler operation state time series data and a sequence of external environment time series data are obtained. After that, the sequence of boiler operation state time series data and the sequence of external environment time series data are input into a time series encoder based on a forward LSTM model to extract and mine the time series features related to the influence of heat load in the boiler operation state and the external environment by utilizing the characteristics of the forward LSTM model that captures long-term time data, and obtain a time series encoding feature vector of the heat load main variable and a time series encoding feature vector of the heat load covariate.

[0038] In an embodiment of the present application, the heat load main-covariate dynamic time series interaction module 140 is used to perform dynamic memory feature inter-attention fusion on the heat load main variable time series coding feature and the heat load covariate time series coding feature to obtain a heat load main-covariate dynamic memory time series interaction fusion representation. It should be understood that there is a time series implicit mutual synergy relationship between the heat load main variable time series coding feature and the heat load covariate time series coding feature. For example, when the outside temperature drops, the heat loss of the building increases, which directly causes the boiler to provide more heat to maintain the indoor temperature. Therefore, in order to more finely capture the interaction between the main variable (boiler operation status data) and the covariate (external environmental information), in particular, this interaction includes not only direct effects, but also indirect or delayed effects. In the technical solution of the present application, the heat load main variable time series coding feature and the heat load covariate time series coding feature are subjected to dynamic memory feature inter-attention fusion to obtain a heat load main-covariate dynamic memory time series interaction fusion representation. In particular, the dynamic memory mechanism can remember past states and use these memories to assist the current decision-making process. This is important for dealing with heat load demands that have significant time dependencies, since past operating states and environmental conditions may have an impact on future demand.

[0039] Specifically, Figure 2 FIG. 1 is a block diagram of a heat load main-covariate dynamic time series interaction module in a hot water boiler intelligent control system based on heat load prediction according to an embodiment of the present application. Figure 2 As shown, the heat load main-covariate dynamic time series interaction module 140 includes: a time series implicit collaborative feature analysis unit 141, which is used to perform collaborative implicit feature extraction on the heat load main variable time series coding feature and the heat load covariate time series coding feature to obtain the heat load main-covariate time series implicit collaborative feature; a feature modulation fusion unit 142, which is used to perform attention modulation fusion on the heat load main variable time series coding feature and the heat load covariate time series coding feature according to the heat load main-covariate time series implicit collaborative feature to obtain the heat load main-covariate dynamic memory time series interaction fusion representation.

[0040] Specifically, in the embodiment of the present application, the time series implicit collaborative feature analysis unit is used to: perform multi-dimensional interactive fusion on the time series encoding feature vector of the heat load main variable and the time series encoding feature vector of the heat load covariate to obtain a heat load main variable-covariate multi-dimensional interactive feature vector. The process can be expressed as: ;in, and represent the time series encoding feature vector of the heat load main variable and the time series encoding feature vector of the heat load covariate, respectively, and They are point-by-point addition, point-by-point subtraction, and point-by-point multiplication. For cascade processing, is the heat load main variable-covariate multidimensional interaction eigenvector.

[0041] The heat load main variable-covariate multidimensional interaction feature vector is subjected to point convolution encoding and Leaky ReLU function activation processing to obtain the heat load main-covariate time series implicit synergy feature vector as the heat load main-covariate time series implicit synergy feature. The process can be expressed as: ;in, is the heat load main variable-covariate multidimensional interaction eigenvector, is the point convolutional encoding operation, is the Leaky ReLU activation function, is the implicit synergistic eigenvector of the heat load principal-covariate time series.

[0042] It should be understood that the collaborative implicit feature extraction of the heat load main variable time-series coding features and the heat load covariate time-series coding features can capture the deep correlation between the heat load main variable time-series coding feature vector and the heat load covariate time-series coding feature vector, that is, the intrinsic, complex and difficult to directly observe correlation between the boiler operating state and external environmental factors. This deep correlation is not limited to simple linear combinations between features, but involves complex nonlinear feature interaction patterns. Specifically, through the point convolution coding operation, the local correlation pattern between the input feature data can be identified, and the use of the Leaky ReLU activation function can ensure the learning of nonlinear feature representation. Such feature data processing operations are conducive to capturing the complex relationship between the heat load main variable time-series coding features and the heat load covariate time-series coding features.

[0043] Specifically, Figure 3 FIG. 1 is a block diagram of a feature modulation fusion unit in a hot water boiler intelligent control system based on heat load prediction according to an embodiment of the present application. Figure 3As shown, the feature modulation fusion unit 142 includes: a heat load main-covariate time series dynamic key vector generation subunit 1421, which is used to write the heat load main-covariate time series implicit collaborative feature vector into a dynamic memory unit to obtain a heat load main-covariate time series dynamic key vector; a heat load main-covariate feature optimization subunit 1422, which is used to dynamically modulate and optimize the heat load main variable time series encoding feature vector and the heat load covariate time series encoding feature vector according to the heat load main-covariate time series dynamic key vector to obtain an optimized heat load main variable time series encoding feature vector and an optimized heat load covariate time series encoding feature vector; a heat load main-covariate feature weighted interaction subunit 1423, which is used to perform feature weighted interaction on the optimized heat load main variable time series encoding feature vector and the optimized heat load covariate time series encoding feature vector to obtain a heat load main-covariate dynamic memory time series interaction fusion vector as the heat load main-covariate dynamic memory time series interaction fusion representation.

[0044] Next, the heat load main-covariate time series implicit cooperative feature vector is written into the dynamic memory unit to obtain the heat load main-covariate time series dynamic key vector. The above process can be expressed as: ;in, is the implicit synergistic eigenvector of the heat load principal-covariate time series, is a 1×1 convolution operation, is the heat load primary-covariate time series dynamic key vector.

[0045] It should be understood that the dynamic memory unit can dynamically adjust its own memory content according to the current input state. By writing the heat load main-covariate time series implicit collaborative feature vector into the dynamic memory unit, the system can better adapt to the dynamic changes of heat load and its related factors. For example, when the external environment suddenly becomes cold (covariate changes), the boiler operating state (main variable) will also be adjusted accordingly. The dynamic memory unit can dynamically update the memory according to this new input, and then provide more realistic information for the subsequent decoded output of the heat load demand result, so as to improve the accuracy of the heat load demand prediction.

[0046] More specifically, in the embodiment of the present application, the heat load main-covariate feature optimization subunit is used to extract the heat load main-covariate time series dynamic key vector from the dynamic memory unit, and perform feature attention optimization based on the first converter on the heat load main variable time series encoding feature vector and the heat load main-covariate time series dynamic key vector to obtain the optimized heat load main variable time series encoding feature vector. This process can be expressed as: ;in, represents the time series encoding feature vector of the main variable of the heat load, is the heat load main variable query embedding matrix, is the first heat load main variable bias vector, is matrix multiplication, is the heat load main variable query vector, is the heat load main variable value embedding matrix, is the second heat load main variable bias vector, is the heat load primary variable value vector, yes The transposed vector of yes Length, yes function, It is the time series encoding feature vector of the main variable of heat load after optimization.

[0047] The heat load main-covariate time series dynamic key vector is extracted from the dynamic memory unit, and the heat load covariate time series encoding feature vector and the heat load main-covariate time series dynamic key vector are subjected to feature attention optimization based on the second converter to obtain the optimized heat load covariate time series encoding feature vector. The process can be expressed as: ;in, represents the time series encoding feature vector of the heat load covariate, is the heat load covariate query embedding matrix, is the first heat load covariate bias vector, is matrix multiplication, is the heat load covariate query vector, is the heat load covariate value embedding matrix, is the second heat load covariate bias vector, is the vector of heat load covariate values, yes function, Time series encoding feature vector of heat load covariates after optimization.

[0048] It should be understood that the heat load main-covariate time series dynamic key vector is extracted from the dynamic memory unit, and the heat load main variable time series encoding feature vector and the heat load main-covariate time series dynamic key vector are optimized based on the first converter to obtain the optimized heat load main variable time series encoding feature vector, and at the same time, the heat load main-covariate time series dynamic key vector is extracted from the dynamic memory unit, and the heat load covariate time series encoding feature vector and the heat load main-covariate time series dynamic key vector are optimized based on the second converter to obtain the optimized heat load covariate time series encoding feature vector. Here, by using the self-attention mechanism of the first converter and the second converter core, the influence of the input heat load main-covariate time series dynamic key vector on the heat load main variable time series encoding feature vector and the heat load covariate time series encoding feature vector at different time steps can be captured respectively, which helps to more comprehensively reflect the law of the original boiler operation state and external environment information changing over time, and is conducive to improving the accuracy and completeness of the corresponding feature representation. Moreover, using the heat load main-covariate time series dynamic key vector as the common attention space not only enhances the expression of key information in the optimized feature vector and reduces the influence of noise and irrelevant information, but also aligns the optimized heat load main variable time series encoding feature vector and the optimized heat load covariate time series encoding feature vector in the semantic space, thereby facilitating the subsequent feature interaction and fusion between the two.

[0049] More specifically, in the embodiment of the present application, the heat load main-covariate feature weighted interaction subunit is used to: calculate the weighted sum between the optimized heat load main variable time series encoding feature vector and the optimized heat load covariate time series encoding feature vector to obtain the heat load main-covariate dynamic memory time series interaction fusion vector, wherein the sum of the weighted coefficient of the optimized heat load main variable time series encoding feature vector and the weighted coefficient of the optimized heat load covariate time series encoding feature vector is one. The above process can be expressed as: ;in, is the time series encoding feature vector of the main variable of heat load after optimization, is the optimized heat load covariate time series encoding feature vector, is the weighting coefficient, It is the heat load main-covariate dynamic memory time series interaction fusion vector.

[0050] It should be understood that by performing feature weighted interaction on the optimized heat load main variable time series encoding feature vector and the optimized heat load covariate time series encoding feature vector, the two optimized feature vectors can be further integrated through linear transformation to generate a more representative heat load main-covariate dynamic memory time series interaction fusion vector.

[0051] In the embodiment of the present application, the boiler fuel supply adjustment module 150 is used to adjust the boiler fuel supply according to the dynamic memory time series interactive fusion representation of the heat load main-covariate. Specifically, Figure 4 FIG. 1 is a block diagram of a boiler fuel supply adjustment module in a hot water boiler intelligent control system based on heat load prediction according to an embodiment of the present application. Figure 4 As shown, the boiler fuel supply adjustment module 150 includes: a heat load demand estimation unit 151, which is used to obtain an estimated value of the heat load demand according to the heat load main-covariate dynamic memory time series interaction fusion vector; a boiler fuel supply adjustment unit 152, which is used to adjust the boiler fuel supply based on the estimated value of the heat load demand.

[0052] In an embodiment of the present application, the heat load demand estimation unit 151 is used to obtain an estimated value of the heat load demand based on the heat load main-covariate dynamic memory time series interaction fusion vector. Specifically, in an embodiment of the present application, the heat load demand estimation unit is used to: input the heat load main-covariate dynamic memory time series interaction fusion vector into a heat load prediction module based on an RNN model to obtain an estimated value of the heat load demand. That is, prediction is performed using the heat load main-covariate dynamic memory time series interaction fusion representation obtained by dynamic memory fusion of the heat load main variable time series encoding feature vector and the heat load covariate time series encoding feature vector, so as to intelligently estimate the heat load demand. In detail, the RNN model is mainly composed of an input layer, a hidden layer, and an output layer. The input layer receives the dynamic memory time series interaction fusion vector of the heat load principal-covariate; the hidden layer is the core part of the RNN, which contains multiple neurons, which are interconnected through recurrent connections. At each time step, the neurons in the hidden layer will update their own states according to the current input and the hidden state of the previous moment, thereby realizing the memory and processing of time series information; the output layer generates an estimated value of the heat load demand based on the final state of the hidden layer. During the training phase, the model will learn the patterns in the data based on a large amount of historical data (including boiler operation status data and external environment information and the corresponding actual heat load demand). Through multiple iterative learning, when a new fusion vector is received, the RNN model will automatically infer on the input data based on the previously learned patterns to predict the future heat load demand. The estimated value of the heat load demand predicted by the RNN model can let the system know the changing trend of the heat load demand, and then adjust the fuel supply of the boiler in advance, so as to achieve efficient, energy-saving and stable operation of the hot water boiler.

[0053] In particular, here, considering that the heat load main variable time series coding feature vector and the heat load covariate time series coding feature vector represent the time series coding features of the boiler operation status data and the external environment information respectively, when performing attention interaction between features based on the dynamic memory mechanism, the insufficient correspondence of the dynamic memory mechanism of the time series coding features under different data source time series will lead to the richness of the attention interaction of the heat load main-covariate dynamic memory time series interaction fusion vector, resulting in repeated unstructured probability mapping, thereby affecting the accuracy of the estimated value of the heat load demand obtained through the heat load prediction module based on the RNN model.

[0054] Preferably, inputting the heat load main-covariate dynamic memory time series interaction fusion vector into the heat load prediction module based on the RNN model to obtain the estimated value of the heat load demand includes: determining the number of zero eigenvalues ​​in the heat load main-covariate dynamic memory time series interaction fusion vector and zero eigenvalue , and the zero eigenfactor value and the number of zero eigenvalues Multiply and divide them respectively to obtain the weight value of the first heat load main-covariate dynamic memory time series interactive fusion field and the second heat load main-covariate dynamic memory time series interactive fusion field weight value .

[0055] Calculate the square root of the sum of squares of all eigenvalues ​​of the heat load main-covariate dynamic memory time series interactive fusion vector to obtain the heat load main-covariate dynamic memory time series interactive fusion full mode weight ;in, Represents each eigenvalue of the heat load main-covariate dynamic memory time series interaction fusion vector.

[0056] With the zero characteristic factor value Calculate each eigenvalue of the heat load main-covariate dynamic memory time series interaction fusion vector as an index The power function , and multiplied by the weight value of the first heat load main-covariate dynamic memory time series interactive fusion field To obtain the first heat load main-covariate dynamic memory time series interactive fusion intermediate value .

[0057] Calculate each eigenvalue of the heat load main-covariate dynamic memory time series interaction fusion vector Dynamic memory time series interaction fusion field weight value with the second heat load main-covariate and the heat load main-covariate dynamic memory time series interactive fusion full mode weight The product of the second heat load main-covariate dynamic memory time series interactive fusion intermediate value is obtained .

[0058] The first heat load main-covariate dynamic memory time series interactive fusion intermediate value Interactive fusion of intermediate values ​​with the second heat load main-covariate dynamic memory time series The weighted difference The optimized heat load main-covariate dynamic memory time series interaction fusion vector is composed of: and They represent different weight hyperparameters respectively.

[0059] The optimized heat load main-covariate dynamic memory time series interaction fusion vector is input into a heat load prediction module based on the RNN model to obtain an estimated value of the heat load demand.

[0060] Accordingly, based on the vector field fitting law of high-dimensional manifold, the isolated zeros and their differentiable factors in the heat load main-covariate dynamic memory time series interaction fusion vector are used as the differentiable dimensions of the high-dimensional manifold, so as to fix the normal direction pointing to the center of the characteristic distribution near the local coordinates represented by the eigenvalues ​​of the heat load main-covariate dynamic memory time series interaction fusion vector, thereby achieving effective alignment of the characteristic distribution pattern of the heat load main-covariate dynamic memory time series interaction fusion vector to the probability density differentiable field to avoid the invalid repetition in the probability mapping process caused by the unstructured characteristics of the distribution of the heat load main-covariate dynamic memory time series interaction fusion vector with rich interactive response, and improve the accuracy of the estimated value of the heat load demand obtained by the heat load main-covariate dynamic memory time series interaction fusion vector through the heat load prediction module based on the RNN model.

[0061] In the embodiment of the present application, the boiler fuel supply adjustment unit 152 is used to adjust the boiler fuel supply based on the estimated value of the heat load demand. It should be understood that during the operation of the hot water boiler, the control system will obtain the estimated value of the heat load demand in real time and compare it with the actual heat output generated by the boiler under the current fuel supply. If the estimated heat load demand is higher than the actual heat output, it means that the fuel supply is insufficient, and the system will increase the fuel supply according to the pre-set control strategy; conversely, if the estimated heat load demand is lower than the actual heat output, the system will reduce the fuel supply. In this way, by accurately adjusting the fuel supply according to the estimated value of the heat load demand, it can be ensured that the boiler can operate in the most optimal state under various working conditions. When the heat load demand is low, energy waste caused by excessive fuel supply can be avoided; and when the heat load demand is high, sufficient heat can be provided in time to ensure the normal operation of the boiler system. At the same time, since the fuel supply can be accurately adjusted according to the heat load demand, excessive combustion of fuel is avoided, thereby reducing the emission of greenhouse gases and harmful pollutants, which helps to reduce the negative impact of the boiler system on the environment.

[0062] Adjusting the boiler fuel supply based on the estimated value of the heat load demand directly determines whether the hot water boiler system can maintain the optimal operating state under various operating conditions. The following is a detailed description of its implementation process: First, real-time monitoring is performed during the operation of the hot water boiler. The estimated value of the heat load demand is continuously obtained. This estimated value is obtained through a series of complex processing of the boiler operating status data and external environmental information. It accurately reflects the actual heat load required by the boiler at the current moment. For example, the latest estimated value of the heat load demand is obtained from the heat load demand estimation unit at regular intervals (such as 5 seconds), and it is used as the key basis for subsequent adjustment of the fuel supply.

[0063] Next, calculate the fuel supply adjustment value. Strictly compare the real-time estimated heat load demand with the actual heat output generated by the current boiler, and calculate the deviation between the two. The calculation of heat output needs to comprehensively consider the steam production or hot water supply of the boiler, and combine factors such as the specific heat capacity and temperature change of water. The deviation calculation formula is ΔQ=Q_est-Q_act, where ΔQ represents the deviation, Q_est is the estimated heat load demand, and Q_act is the actual heat output of the boiler. For example, if the estimated heat load demand is 800kW, and the actual heat output of the boiler is calculated to be 700kW, the deviation is 100kW, which indicates that the current fuel supply is insufficient and needs to be increased. Based on the established heat load demand and fuel supply relationship model and the calculated deviation, calculate the fuel supply adjustment value. If a linear regression model is used, the adjustment value calculation formula is ΔF=ΔQ / a (where ΔF is the fuel supply adjustment value). For example, if a=0.7 and deviation ΔQ=100kW, then the fuel supply adjustment value ΔF=100 / 0.7≈142.86kg / h, which means that the fuel supply needs to be increased by about 142.86kg / h. For neural network models or other complex models, the adjustment value is calculated automatically by the model based on the input deviation and other related parameters. The model outputs the appropriate fuel supply adjustment value with its powerful learning ability and complex calculation logic.

[0064] Then, the fuel supply amount is adjusted. The control system generates corresponding control instructions based on the calculated fuel supply amount adjustment value. These control instructions can be digital signals or analog signals, and their purpose is to accurately control various regulating devices in the fuel supply system, such as valve opening, pump speed, etc. For example, if the calculation results show that the fuel supply amount needs to be increased, the control system will generate a signal to increase the fuel supply valve opening by a certain angle, or increase the speed of the fuel pump by a certain proportion. Subsequently, the generated control instructions are quickly sent to the actuators of the fuel supply system, such as electric control valves and frequency converters. After receiving the control signal, the electric control valve is driven by a motor to accurately change the valve opening, thereby accurately controlling the flow of fuel; the frequency converter changes the power supply frequency of the fuel pump according to the control signal, and then adjusts the speed of the pump to achieve fine adjustment of the fuel supply amount. During the execution process, the accuracy and timeliness of the actuator action must be ensured, and the actual changes such as fuel flow rate are monitored in real time through sensors, and feedback monitoring of the execution results is performed to determine whether the adjustment has achieved the expected effect.

[0065] Finally, feedback and optimization adjustments are made. After the fuel supply is adjusted, the operating parameters and heat output of the boiler are closely monitored to comprehensively evaluate the adjustment effect. Carefully observe whether the parameters such as the water temperature and pressure of the boiler change in the expected direction, and whether the heat output truly meets the heat load demand. For example, if the boiler water temperature is monitored to rise steadily and within a reasonable range after increasing the fuel supply, it means that the adjustment effect is good; on the contrary, if the water temperature rises too fast, it may cause the pressure to be too high, which indicates that there may be problems with the adjustment and the adjustment strategy needs to be further optimized. According to the evaluation results of the adjustment effect, the adjustment strategy is optimized in a targeted manner. If it is found that the adjusted fuel supply still cannot meet the heat load demand, or it causes problems such as unstable boiler operation, it is necessary to analyze the reasons in depth. It may be that the established relationship model is not accurate enough and fails to fully consider certain special working conditions or large changes in the external environment; it may also be that some parameters are set unreasonably when calculating the adjustment value. For these problems, the model can be retrained or revised, the parameters for calculating the adjustment value can be readjusted, or the model can be further improved to increase the consideration of other factors (such as the impact of changes in fuel quality). The fuel supply adjustment value is then calculated again and the adjustment is performed. Through this continuous feedback and optimization cycle, the adjustment of the boiler fuel supply can be made more accurate and reasonable, ensuring that the boiler always operates in the optimal state.

[0066] In summary, the hot water boiler intelligent control system 100 based on heat load prediction according to the embodiment of the present application is explained, which uses artificial intelligence-based data analysis and processing methods to use the boiler operation status data collected by the sensor matrix as the main variable and the external environment information as the covariate, and then extracts the heat load main variable time series coding features and the heat load covariate time series coding features from the main variable and the covariate respectively, so as to realize intelligent estimation of heat load demand based on the dynamic memory interaction representation between the heat load main variable time series coding features and the heat load covariate time series coding features and adjust the boiler fuel supply in real time accordingly. In this way, it is possible to respond to changes in the external environment in real time, ensure that the boiler is always operating in the optimal state, help reduce energy waste and unnecessary emissions, and maintain an efficient operating state in a complex and changing environment.

[0067] Figure 5 FIG. 1 is a flow chart of a hot water boiler intelligent control method based on heat load prediction according to an embodiment of the present application. Figure 5As shown, in the hot water boiler intelligent control method based on heat load prediction, it includes: S110, using a sensor matrix to collect boiler operation status data; S120, collecting external environment information; S130, taking the boiler operation status data as the main variable and the external environment information as the covariate, extracting the heat load main variable time series coding features and the heat load covariate time series coding features from the main variable and the covariate; S140, performing dynamic memory feature inter-attention fusion on the heat load main variable time series coding features and the heat load covariate time series coding features to obtain the heat load main-covariate dynamic memory time series Interactive fusion representation; S150, according to the thermal load main-covariate dynamic memory time series interactive fusion representation, adjust the boiler fuel supply, wherein the step S140 includes: extracting collaborative implicit features of the thermal load main variable time series coding features and the thermal load covariate time series coding features to obtain the thermal load main-covariate time series implicit collaborative features; according to the thermal load main-covariate time series implicit collaborative features, performing attention modulation fusion on the thermal load main variable time series coding features and the thermal load covariate time series coding features to obtain the thermal load main-covariate dynamic memory time series interactive fusion representation.

[0068] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned hot water boiler intelligent control method based on heat load prediction have been referred to above. Figures 1 to 4 The description of the hot water boiler intelligent control system based on heat load prediction has been introduced in detail, and therefore, its repeated description will be omitted.

[0069] In summary, the hot water boiler intelligent control method based on heat load prediction based on the embodiment of the present application is explained, which uses artificial intelligence-based data analysis and processing methods to take the boiler operation status data collected by the sensor matrix as the main variable and the external environment information as the covariate, and then extracts the heat load main variable time series coding characteristics and the heat load covariate time series coding characteristics from the main variable and the covariate respectively, so as to realize the intelligent estimation of heat load demand based on the dynamic memory interaction between the heat load main variable time series coding characteristics and the heat load covariate time series coding characteristics and adjust the boiler fuel supply in real time accordingly. In this way, it is possible to respond to changes in the external environment in real time, ensure that the boiler is always operating in the optimal state, help reduce energy waste and unnecessary emissions, and maintain an efficient operating state in a complex and changing environment.

Claims

1. An intelligent control system for hot water boilers based on heat load prediction, characterized in that: include: A boiler operation status data acquisition module is used to collect boiler operation status data using a sensor matrix; An external environment information acquisition module is used to collect external environment information; A data time series coding module, used to take the boiler operation status data as a main variable and the external environment information as a covariate, and extract a heat load main variable time series coding feature and a heat load covariate time series coding feature from the main variable and the covariate; The heat load main-covariate dynamic time series interaction module is used to perform dynamic memory feature attention fusion on the heat load main variable time series coding feature and the heat load covariate time series coding feature to obtain the heat load main-covariate dynamic memory time series interaction fusion representation, wherein the heat load main-covariate dynamic time series interaction module includes: a time series implicit collaborative feature analysis unit, used to perform collaborative implicit feature extraction on the heat load main variable time series coding feature and the heat load covariate time series coding feature to obtain the heat load main-covariate time series implicit collaborative feature; a feature modulation fusion unit, used to perform attention modulation fusion on the heat load main variable time series coding feature and the heat load covariate time series coding feature according to the heat load main-covariate time series implicit collaborative feature to obtain the heat load main-covariate dynamic memory time series interaction fusion representation; a boiler fuel supply adjustment module, used to adjust the boiler fuel supply according to the heat load main-covariate dynamic memory time series interaction fusion representation.

2. The hot water boiler intelligent control system based on heat load prediction according to claim 1 is characterized in that: The data time series encoding module includes: a data integration unit, which is used to integrate the boiler operation status data and the external environment information according to the time dimension to obtain a sequence of boiler operation status time series data and a sequence of external environment time series data; a data time series encoding feature generation unit, which is used to input the sequence of boiler operation status time series data and the sequence of external environment time series data into a time series encoder based on a forward LSTM model to obtain a heat load main variable time series encoding feature vector as the heat load main variable time series encoding feature and a heat load covariate time series encoding feature vector as the heat load covariate time series encoding feature.

3. The hot water boiler intelligent control system based on heat load prediction according to claim 2 is characterized in that: The time series implicit collaborative feature analysis unit is used to: perform multi-dimensional interactive fusion on the time series encoding feature vector of the heat load main variable and the time series encoding feature vector of the heat load covariate to obtain a heat load main variable-covariate multi-dimensional interactive feature vector; perform point convolution encoding and Leaky ReLU function activation processing on the heat load main variable-covariate multi-dimensional interactive feature vector to obtain a heat load main-covariate time series implicit collaborative feature vector as the heat load main-covariate time series implicit collaborative feature.

4. The hot water boiler intelligent control system based on heat load prediction according to claim 3 is characterized in that: The feature modulation and fusion unit includes: a heat load main-covariate time series dynamic key vector generation subunit, which is used to write the heat load main-covariate time series implicit collaborative feature vector into a dynamic memory unit to obtain a heat load main-covariate time series dynamic key vector; a heat load main-covariate feature optimization subunit, which is used to dynamically modulate and optimize the heat load main variable time series encoding feature vector and the heat load covariate time series encoding feature vector according to the heat load main-covariate time series dynamic key vector to obtain an optimized heat load main variable time series encoding feature vector and an optimized heat load covariate time series encoding feature vector; a heat load main-covariate feature weighted interaction subunit, which is used to perform feature weighted interaction on the optimized heat load main variable time series encoding feature vector and the optimized heat load covariate time series encoding feature vector to obtain a heat load main-covariate dynamic memory time series interaction fusion vector as the heat load main-covariate dynamic memory time series interaction fusion representation.

5. The hot water boiler intelligent control system based on heat load prediction according to claim 4 is characterized in that: The heat load main-covariate feature optimization subunit is used to: extract the heat load main-covariate time series dynamic key vector from the dynamic memory unit, and perform feature attention optimization based on the first converter on the heat load main variable time series encoding feature vector and the heat load main-covariate time series dynamic key vector to obtain the optimized heat load main variable time series encoding feature vector; The heat load main-covariate time series dynamic key vector is extracted from the dynamic memory unit, and the heat load covariate time series encoding feature vector and the heat load main-covariate time series dynamic key vector are subjected to feature attention optimization based on a second converter to obtain the optimized heat load covariate time series encoding feature vector.

6. The hot water boiler intelligent control system based on heat load prediction according to claim 5 is characterized in that: The heat load main-covariate feature weighted interaction subunit is used to calculate the weighted sum between the optimized heat load main variable time series encoding feature vector and the optimized heat load covariate time series encoding feature vector to obtain the heat load main-covariate dynamic memory time series interaction fusion vector, wherein the sum of the weighted coefficient of the optimized heat load main variable time series encoding feature vector and the weighted coefficient of the optimized heat load covariate time series encoding feature vector is one.

7. The hot water boiler intelligent control system based on heat load prediction according to claim 6 is characterized in that: The boiler fuel supply adjustment module includes: a heat load demand estimation unit, which is used to obtain an estimated value of the heat load demand according to the heat load main-covariate dynamic memory time series interactive fusion vector; a boiler fuel supply adjustment unit, which is used to adjust the boiler fuel supply based on the estimated value of the heat load demand.

8. The hot water boiler intelligent control system based on heat load prediction according to claim 7 is characterized in that: The heat load demand estimation unit is used to: input the heat load main-covariate dynamic memory time series interaction fusion vector into the heat load prediction module based on the RNN model to obtain the estimated value of the heat load demand.

9. A hot water boiler intelligent control method based on heat load prediction, characterized in that: include: Use sensor matrix to collect boiler operation status data; Collect external environmental information; Taking the boiler operation status data as a main variable and the external environment information as a covariate, extracting a heat load main variable time series coding feature and a heat load covariate time series coding feature from the main variable and the covariate; Performing dynamic memory feature inter-attention fusion on the time series coding features of the heat load main variable and the time series coding features of the heat load covariate to obtain a heat load main-covariate dynamic memory time series interactive fusion representation, including: performing collaborative implicit feature extraction on the time series coding features of the heat load main variable and the time series coding features of the heat load covariate to obtain a heat load main-covariate time series implicit collaborative feature; According to the implicit collaborative characteristics of the heat load main-covariate time series, the attention modulation fusion is performed on the heat load main variable time series coding characteristics and the heat load covariate time series coding characteristics to obtain the heat load main-covariate dynamic memory time series interactive fusion representation; according to the heat load main-covariate dynamic memory time series interactive fusion representation, the boiler fuel supply is adjusted.

10. The hot water boiler intelligent control method based on heat load prediction according to claim 9, characterized in that: The method uses the boiler operation state data as a main variable and the external environment information as a covariate, and extracts the heat load main variable time series coding features and the heat load covariate time series coding features from the main variable and the covariate, including: integrating the boiler operation state data and the external environment information according to the time dimension to obtain a sequence of boiler operation state time series data and a sequence of external environment time series data; The sequence of the boiler operation status time series data and the sequence of the external environment time series data are input into a time series encoder based on a forward LSTM model to obtain a heat load main variable time series encoding feature vector as the heat load main variable time series encoding feature and a heat load covariate time series encoding feature vector as the heat load covariate time series encoding feature.

Images