A centralized heating network intelligent self-organizing production scheduling method and related device

Abstract

The present application belongs to the technical field of automation control and intelligent optimization scheduling of central heating system, and discloses a kind of central heating network intelligent self-organizing production scheduling method and related device, comprising: obtaining the historical time series data of heat source and each heat exchange station, calculating the transmission entropy of heat source to each heat exchange station, determining the dynamic delivery time lag of each heat exchange station by maximizing transmission entropy, constructing full-network dynamic causal connectivity graph, constructing total loss function containing data error, mass conservation residual and energy conservation residual for training, obtaining physical information neural network;With the sum of heat source heating energy consumption and main circulating pump delivery power consumption being the lowest as the target, the physical information neural network is embedded into the optimization algorithm as a dynamic implicit constraint, and combined with the dynamic delivery time lag of each heat exchange station, the required heat source water temperature and total flow are output as scheduling instructions.The present application can realize safe and reliable central heating intelligent scheduling without pipe network drawing and physical mechanism model.

Description

Technical Field

[0001] This invention belongs to the field of automated control and intelligent optimization scheduling technology of centralized heating systems, and relates to a method and related devices for intelligent self-organized production scheduling of centralized heating networks. Background Technology

[0002] As a crucial urban infrastructure, the accuracy and reliability of centralized heating systems directly impact the heating quality and system operational safety for tens of thousands of users. Traditional smart heating production scheduling strategies heavily rely on precise Geographic Information System (GIS) topology maps and detailed physical parameters such as pipe diameter, pipe length, and insulation thickness. This data is used to establish complex sets of hydraulic and thermodynamic partial differential equations for joint optimization calculations of hydraulic and thermal conditions. Centralized heating primary pipe networks typically involve long transmission distances, complex topologies, and large coverage areas. The corresponding nonlinear partial differential equation models require numerous physical parameters such as pipe length, diameter, and friction coefficients, resulting in a high engineering threshold for the modeling process itself.

[0003] However, in actual engineering projects, many old heating pipe networks were built a long time ago. The actual routes, connections, and pipe diameter parameters of underground pipelines often suffer from serious data gaps or topological errors due to missing historical drawings, poor record management, or failure to update data in a timely manner during renovations. In urban natural gas pipeline networks, heating pipeline networks, and other underground pipeline systems, GIS data commonly suffers from missing pipeline information, inaccurate burial depth, and unclear routes. Under such network conditions with unclear topology and parameters, any scheduling strategy relying on physical mechanism models completely fails due to distorted input parameters, making traditional modeling methods difficult to implement.

[0004] On the other hand, completely abandoning physical mechanism models and directly adopting purely data-driven traditional deep learning networks (such as conventional recurrent neural networks (RNNs) and long short-term memory networks (LSTMs) as a "black box" alternative faces another serious technical problem. Purely data-driven methods generally cannot utilize any prior physical knowledge and ignore fundamental physical phenomena such as the conservation of mass and energy. Because sensor data in industrial settings inevitably contains measurement noise, communication packet loss, and abnormal fluctuations, purely data-driven models are prone to fitting noise during the learning process, leading to physically invalid solutions that violate the first and second laws of thermodynamics, such as energy conservation and return water temperatures exceeding supply water temperatures. Once such scheduling instructions that violate basic thermodynamic laws are issued and executed, they will pose significant safety hazards to heating production and operation, potentially causing hydraulic instability or equipment damage in severe cases.

[0005] Currently, some research has attempted to apply Physical Information Neural Networks (PINNs) to the modeling and predictive control of heating systems. This involves embedding partial differential equation constraints into the loss function of the neural network to achieve a fusion of data-driven approaches and physical laws. However, existing PINN methods typically still rely on known pipeline topologies when constructing physical constraints, embedding the physical topology into the network interconnections, or requiring pre-obtained pipeline topology connections as graph input. In other words, while existing PINN schemes introduce physical constraints at the model training level, their application still presupposes the availability of accurate and usable pipeline topology maps, failing to fundamentally solve the heating scheduling problem in scenarios where "no topology is available." When the pipeline topology is unknown or data is missing, existing PINN methods are also difficult to deploy and implement.

[0006] In summary, existing centralized heating production scheduling technologies have the following shortcomings: (1) methods relying on precise physical mechanism models cannot be implemented under conditions where the topology of old pipe networks is missing; (2) purely data-driven "black box" models are prone to generating invalid solutions that violate the laws of thermodynamics and physics, posing safety hazards; (3) existing neural network methods that integrate physical information still rely on known pipe network topology structures and have not truly broken free from dependence on drawings and mechanism models. Therefore, there is an urgent need in this field for an intelligent production scheduling method for centralized heating networks that does not rely on pipe network drawings and physical mechanism models, while satisfying the constraints of thermodynamic and physical laws. Summary of the Invention

[0007] The purpose of this invention is to provide a method and related device for intelligent self-organizing production scheduling of centralized heating networks, which solves the problem that existing technologies cannot be implemented or are prone to violating thermodynamic and physical laws when there is a lack of accurate pipeline topology drawings and physical mechanism models.

[0008] To achieve the above objectives, the present invention employs the following technical solution: A method for intelligent self-organizing production scheduling in centralized heating networks includes: Historical time-series operation data of heat sources and each heat exchange station are obtained. The transmission entropy of heat source to each heat exchange station is calculated. The dynamic transmission and distribution time delay that maximizes the transmission entropy is determined by adjusting the time lag order. The dynamic causal connectivity graph of the whole network is obtained, and the dynamic transmission and distribution time delay of each heat exchange station is obtained. A long short-term memory network is constructed, using the dynamic transmission and distribution time delay of each heat exchange station. The current heat source output data and the historical state data of each heat exchange station under their respective dynamic transmission and distribution time delay are used as inputs, and the predicted temperature field and predicted pressure field of each heat exchange station at future time are used as outputs. A total loss function including data error loss, continuous mass conservation residual and energy conservation and heat dissipation attenuation physical residual is constructed for training to obtain a physical information neural network. The objective function is constructed with the goal of minimizing the sum of the heat source heating energy consumption and the main circulation pump power consumption. The physical information neural network is embedded as a dynamic implicit constraint function into the optimization algorithm. The candidate scheduling scheme that minimizes the objective function value is selected as the optimal scheduling scheme, and the corresponding heat source water supply temperature and total heat source flow rate are output as scheduling instructions.

[0009] Furthermore, the heat transfer entropy from the heat source to each heat exchange station is:

[0010] in, Indicates the heat source. Indicates the first A heat exchange station, Indicates heat source to heat exchange station Transmission entropy, Indicates the time of the heat source water supply temperature, Indicates the first Each heat exchange station at time The inlet water temperature, Denotes the joint probability density. and These represent the time lag order, respectively.

[0011] Furthermore, the data error loss is:

[0012] in, Indicates the number of samples. Indicates the actual temperature. Indicates the predicted temperature; The residual for continuous mass conservation is:

[0013] in, This represents the node index of the heat exchange station. Indicates the inflow of the first Total flow rate of each heat exchange station node Indicates the outflow of the first Total flow rate of each heat exchange station node; The physical residuals of energy conservation and heat dissipation decay are:

[0014] in, Indicates the node index. Indicates the first Predicted temperature of each node, Indicates time, Indicates the first The flow rate of hot water at each node Indicates the first Predict the spatial gradient of temperature at each node. Indicates the first The heat loss coefficient of the pipe at each node, This indicates the density of hot water. This indicates the specific heat capacity of hot water. This indicates the temperature of the underground soil.

[0015] Furthermore, the total loss function is:

[0016] in, Indicates data error loss. This represents the residual due to continuous mass conservation. This represents the physical residuals due to energy conservation and heat dissipation decay. and They represent and Weighting coefficients in the total loss function.

[0017] Furthermore, the objective function is:

[0018] in, Represents the objective function value. This represents the unit heat cost of natural gas, coal, or industrial waste heat consumed by the heat source. Indicates time Total flow rate of the main circulation pump Indicates time The temperature of the heat source water supply, Indicates time The return water temperature of the heat source, The time-of-use electricity price indicates the amount of electricity consumed by the main circulation pump. Indicates time The main circulation pump pressure differential.

[0019] Furthermore, the physical information neural network, as a dynamic implicit constraint function, has the following input-output relationship:

[0020] in, This indicates the predicted temperature for each heat exchange station. This represents a dynamic time delay matrix composed of the dynamic transmission and distribution time delays of all heat exchange stations in the entire network. Represents a neural network of physical information. Indicates time The temperature of the heat source water supply, Indicates time Total flow rate of the main circulation pump.

[0021] Furthermore, the process of selecting the optimal scheduling scheme by the optimization algorithm is as follows: Each iteration attempts different heat source water supply temperatures and total heat source flow rates, and the predicted temperature of each heat exchange station is output by the physical information neural network. The schemes that satisfy the condition that the predicted temperature of each heat exchange station is not lower than its own minimum comfort temperature threshold are selected as candidate scheduling schemes, and the candidate scheduling scheme that minimizes the objective function value is selected as the optimal scheduling scheme. The constraint is that the predicted temperature of each heat exchange station is not lower than its respective minimum comfort temperature threshold.

[0022] in, This indicates the predicted temperature for each heat exchange station. This represents a dynamic time delay matrix composed of the dynamic transmission and distribution time delays of all heat exchange stations in the entire network. This indicates the minimum comfort temperature threshold for each heat exchange station.

[0023] A centralized heating network intelligent self-organizing production scheduling system includes: The calculation module is used to acquire historical time-series operation data of heat sources and each heat exchange station, calculate the transmission entropy of heat source to each heat exchange station, determine the dynamic transmission and distribution time delay that maximizes the transmission entropy by adjusting the time lag order, obtain the dynamic causal connectivity graph of the whole network, and obtain the dynamic transmission and distribution time delay of each heat exchange station. The modeling module is used to construct a long short-term memory network. It adopts the dynamic transmission and distribution time delay of each heat exchange station, takes the current heat source output data and the historical state data of each heat exchange station under its own dynamic transmission and distribution time delay as input, and takes the predicted temperature field and predicted pressure field of each heat exchange station at the future time as output. It constructs a total loss function including data error loss, continuous mass conservation residual and energy conservation and heat dissipation attenuation physical residual for training, and obtains physical information neural network. The optimization module is used to construct an objective function with the goal of minimizing the sum of the heating energy consumption of the heat source and the power consumption of the main circulation pump. It embeds a physical information neural network as a dynamic implicit constraint function into the optimization algorithm, selects the candidate scheduling scheme that minimizes the objective function value as the optimal scheduling scheme, and outputs the corresponding heat source water supply temperature and total heat source flow rate as scheduling instructions.

[0024] An electronic device includes 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 the steps of the method.

[0025] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method.

[0026] Compared with the prior art, the present invention has the following beneficial effects: This invention provides an intelligent self-organizing production scheduling method for centralized heating networks. By calculating the transmission entropy of heat sources to each heat exchange station and maximizing the dynamic distribution delay of each heat exchange station, it reverse-engineers the dynamic causal connectivity graph of the entire network without requiring any physical parameters such as network topology maps, pipe diameters, or pipe lengths from a geographic information system. This enables the effective implementation of traditional smart heating systems in old or frequently renovated networks with unclear topologies and parameters, completely eliminating reliance on underground network maps and significantly reducing the initial engineering costs of system deployment. Secondly, by constructing a long short-term memory network and using the dynamic distribution delay of each heat exchange station as input, along with the current heat source output and the historical state of each heat exchange station, and introducing continuous mass conservation residuals and energy conservation and heat dissipation physical residuals into the total loss function, the first law of thermodynamics is forcibly written into the genes of the neural network, allowing the trained physical information neural network to function even in the absence of machine learning. Under the premise of theoretical model, it spontaneously grasps the thermodynamic flow law of the unknown pipeline network bottom layer, effectively eliminating the invalid solutions that violate thermodynamic physical laws, such as those that do not conform to energy conservation and return water temperature is higher than supply water temperature, which are generated by industrial data noise in pure data-driven models. This significantly improves the safety and reliability of scheduling operation. Finally, in the optimization stage, physical information neural network is embedded as a dynamic implicit constraint function into the optimization algorithm. With the goal of minimizing the sum of heat source heating energy consumption and main circulation pump transmission and distribution power consumption, and combined with the dynamic transmission and distribution time delay of each heat exchange station, the predicted temperature of each heat exchange station is output. Under the premise of meeting user comfort constraints, the optimal scheduling scheme is selected, which shortens the forward prediction calculation from minutes to milliseconds in the traditional partial differential equation system solution. This meets the real-time requirements of high-frequency rolling production scheduling of large-scale urban heating networks and realizes energy-saving, safe and efficient intelligent self-organizing production scheduling of centralized heating without relying on drawings and mechanism models and can guarantee thermodynamic physical law constraints.

[0027] Furthermore, the nonlinear causal influence between the heat source supply water temperature time series and the heat exchange station inlet water temperature time series is quantified into a calculable transmission entropy value. By comparing the ratio of conditional probabilities before and after incorporating historical heat source data, a logarithmic term greater than 0 proves that the heat source temperature change is the "cause" of the heat exchange station temperature change. Thus, without relying on any physical mechanism or topological information, the causal relationship between the heat source and the heat exchange station can be accurately identified solely based on sensor time series data. This provides a reliable foundation for determining the dynamic transmission and distribution time delay and constructing the dynamic causal connectivity graph, ensuring the accuracy and robustness of the inverse topology.

[0028] Furthermore, data error loss ensures that the neural network output is numerically close to the actual observed data, guaranteeing the basic accuracy of the prediction. The continuous mass conservation residual establishes a constraint that the difference between the total inflow and outflow at the heat exchange station nodes is zero, enabling the neural network to learn the fundamental hydraulic law of flow conservation at nodes in a branchless pipe network. The energy conservation and heat dissipation decay physical residuals embed the first law of thermodynamics into the network training process. These three terms correspond to the rate of change of node temperature over time, the convective transfer of heat along the spatial gradient, and the heat loss caused by the temperature difference between the pipe and the soil, respectively. When the sum of these three terms approaches zero, the temperature field output by the neural network satisfies both the physical law of heat dissipation decay along the pipe and ensures the conservation of heat in both time and space, fundamentally preventing the generation of invalid solutions that violate the law of energy conservation.

[0029] Furthermore, the penalty intensity of physical constraints during training can be flexibly adjusted through weight coefficients, enabling the neural network to achieve a balance between fitting data and following physical laws. When the physical residual is not zero, the total loss function generates a huge penalty term, forcing the optimizer to adjust the network weights so that the output approaches the physical feasible region. As training progresses, when the physical residual approaches zero, it means that the neural network has not only learned to fit historical data, but also mastered the thermodynamic flow law of the unknown pipeline network, realizing self-organizing learning that integrates data-driven and physical laws. This not only retains the ability of deep learning to fit complex nonlinear systems, but also ensures the physical rationality of the output.

[0030] Furthermore, the objective function comprehensively considers the fuel or waste heat consumption costs on the heat source side and the electricity consumption costs on the transmission and distribution side, enabling global optimization to achieve overall economic operation of the heating system. The inclusion of time-of-use electricity pricing in the objective function allows the optimization algorithm to automatically adjust its electricity consumption strategy based on peak and off-peak electricity price differences. During off-peak hours, it appropriately increases transmission and distribution flow, while prioritizing thermal inertia to maintain heating during peak hours, achieving economical load scheduling that follows the source and effectively reducing system operating electricity costs. The 24-hour rolling optimization cycle balances the response characteristics of the large-inertia system of the heating network with the trend of outdoor temperature changes, ensuring both the stability of heating quality and providing sufficient operational lead time for the heat source units.

[0031] Furthermore, by embedding the trained physical information neural network as a constraint condition of the objective function into the optimization algorithm, the underlying thermodynamic flow laws it has mastered are solidified in the network weights. Therefore, during each iteration of the optimization process, when trying different heat source water supply temperatures and total flow rates, the physical information neural network can directly output the predicted temperature of each heat exchange station combined with its own causal time lag within milliseconds. This completely replaces the step of solving large-scale partial differential equations online in traditional methods, reducing the calculation time from minutes to milliseconds, making it possible to implement high-frequency rolling optimization scheduling in large-scale urban heating networks.

[0032] Furthermore, by using the lowest secondary network water supply temperature or the equivalent indoor temperature control lower limit in the area under the jurisdiction of each heat exchange station as the constraint boundary, it is ensured that the optimal scheduling scheme obtained through optimization must meet the heating quality requirements of users in the areas under the jurisdiction of all heat exchange stations while pursuing the lowest energy consumption, thus achieving the dual goals of energy saving and supply guarantee. At the same time, since the predicted temperature output is calculated based on the actual dynamic transmission and distribution time delay of each heat exchange station rather than a fixed empirical time delay, the constraint verification is more in line with the actual dynamic characteristics of heat transmission and distribution in the pipeline network, avoiding the problem of undersupply or oversupply caused by time delay estimation deviation, and effectively ensuring the comfort of end users. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart of the intelligent self-organizing production scheduling method for centralized heating networks according to the present invention.

[0035] Figure 2 This is a schematic diagram of the structure of a centralized heating network intelligent self-organizing production scheduling system, which is a preferred embodiment of the present invention.

[0036] Figure 3 This is a schematic diagram of the electronic device structure according to a preferred embodiment of the present invention. Detailed Implementation

[0037] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0038] Obviously, the described embodiments are only some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0039] It should be noted that the terminals involved in the embodiments of this application may include, but are not limited to, mobile phones, personal digital assistants (PDAs), wireless handheld devices, tablet computers, personal computers (PCs), MP3 players, MP4 players, wearable devices (e.g., smart glasses, smartwatches, smart bracelets), smart home devices, and other smart devices.

[0040] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0041] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1 This invention provides an intelligent self-organizing production scheduling method for centralized heating networks, specifically including the following steps: Step 1: Construct a dynamic causal connectivity diagram of the pipeline network.

[0042] Historical time-series operation data of the heat source and each heat exchange station are acquired. Based on the historical time-series operation data, the transmission entropy of the heat source supply water temperature time sequence to the inlet water temperature time sequence of each heat exchange station is calculated. The transmission entropy is used to quantify the nonlinear causal influence of the heat source temperature state on the temperature state of the heat exchange station.

[0043] Specifically, let the heat source water supply temperature sequence be... , No. The actual primary network inlet water temperature sequence of each heat exchange station is as follows: Calculate the heat source using the following formula. to heat exchange station Transmission entropy :

[0044] in, Indicates the heat source. Indicates the first A heat exchange station, Indicates heat source to heat exchange station Transmission entropy, Indicates the time of the heat source water supply temperature, Indicates the first Each heat exchange station at time The inlet water temperature, Denotes the joint probability density. and These represent the time lag order, respectively.

[0045] In the above formula This indicates that it is based only on the heat exchange station. Use its own historical data to predict the temperature at the next moment; This means adding a heat source to the existing historical data of the heat exchange station. Historical data is used to predict the temperature of the heat exchange station at the next moment. If the prediction accuracy is significantly improved after adding heat source data, the logarithm is greater than 0, indicating that the temperature change of the heat source is the "cause" of the temperature change of the heat exchange station.

[0046] By adjusting the time lag order Change the time delay The value of will make the transmission entropy The maximum time delay value As heat flows from the heat source to the first The dynamic transport and distribution time delays of each heat exchange station are calculated. Therefore, without requiring network diagrams, the dynamic transport and distribution time delays of all heat exchange stations in the entire network are calculated and combined into a dynamic time delay matrix. .

[0047] Set a significance threshold ,like In the implicit causal topological graph Establish a directed edge from the heat source to the heat exchange station, and the weight of the edge is the dynamic transmission and distribution time delay of the heat exchange station. By connecting all heat exchange stations that pass the significance test to the heat source, a dynamic causal connectivity graph of the entire network's thermal and hydraulic signals is constructed.

[0048] Step 2: Train the physical information neural network.

[0049] A Long Short-Term Memory (LSTM) network is constructed, taking the current heat source output data and the historical state data of each heat exchange station under its respective dynamic transmission and distribution time lag as inputs to the LSTM network, and outputting the future data after forward propagation. Predicted temperature field at each heat exchange station at any time With predicted pressure field .

[0050] Construct the total loss function The total loss function consists of two parts: data error loss and physical residual loss.

[0051] in, Indicates data error loss. This represents the residual due to continuous mass conservation. This represents the physical residuals due to energy conservation and heat dissipation decay. and They represent and Weighting coefficients in the total loss function.

[0052] The data error loss is the mean square error between the predicted temperature and the actual temperature, calculated according to the following formula:

[0053] in, Indicates the number of samples. Indicates the actual temperature. This indicates the predicted temperature.

[0054] The continuous mass conservation residual, for each heat exchange station node, is calculated using the following formula:

[0055] in, This represents the node index of the heat exchange station. Indicates the inflow of the first Total flow rate of each heat exchange station node Indicates the outflow of the first Total flow rate of each heat exchange station node.

[0056] The energy conservation and heat dissipation attenuation physical residuals are calculated according to the following formula:

[0057] in, Indicates the node index. Indicates the first Predicted temperature of each node, Indicates time, Indicates the first The flow rate of hot water at each node Indicates the first Predict the spatial gradient of temperature at each node. Indicates the first The heat loss coefficient of the pipe at each node, This indicates the density of hot water. This indicates the specific heat capacity of hot water. This indicates the temperature of the underground soil.

[0058] In the above formula This represents the rate of change of temperature at a certain node over time. Indicates the flow rate of hot water The spatial gradient transfer of heat during spatial propulsion; Indicates hot water in the pipes With underground soil When a temperature difference exists, heat is lost outward through the pipe wall. This is the heat dissipation loss coefficient.

[0059] According to the first law of thermodynamics, the sum of the above three terms should physically equal zero. When the temperature predicted by the Long Short-Term Memory network... When the thermodynamic laws apply, the calculated value within the parentheses is 0. When the prediction is inaccurate, the calculated value within the parentheses will be non-zero, resulting in a non-zero value. Penalties.

[0060] Using automatic differentiation techniques, the Long Short-Term Memory (LSTM) network is trained by continuously minimizing the total loss function through an optimizer. When the physical residuals approach zero, a well-trained physical information neural network is obtained. This physical information neural network not only learns to fit data but also grasps the thermodynamic flow patterns of the unknown pipeline network's underlying structure.

[0061] Step 3: Continuously solve for the optimal solution of production scheduling.

[0062] The objective function is constructed with the goal of minimizing the sum of the heating energy consumption of the heat source and the power consumption of the main circulation pump within a preset future time period. Specifically, the objective function is constructed according to the following formula, with the goal of minimizing the sum of the heating energy consumption of the heat source and the power consumption of the main circulation pump within the next 24 hours:

[0063] in, Represents the objective function value. This represents the unit heat cost of natural gas, coal, or industrial waste heat consumed by the heat source. Indicates time Total flow rate of the main circulation pump Indicates time The temperature of the heat source water supply, Indicates time The return water temperature of the heat source, The time-of-use electricity price indicates the amount of electricity consumed by the main circulation pump. Indicates time The main circulation pump pressure differential.

[0064] The trained physical information neural network is embedded as a dynamic implicit constraint function in the optimization algorithm, and its input-output relationship is as follows:

[0065] in, This indicates the predicted temperature for each heat exchange station. This represents a dynamic time delay matrix composed of the dynamic transmission and distribution time delays of all heat exchange stations in the entire network. This represents the physical information neural network. Indicates time The temperature of the heat source water supply, Indicates time Total flow rate of the main circulation pump.

[0066] The optimization algorithm tries different heat source supply water temperatures and total heat source flow rates in each iteration, inputs the tried heat source supply water temperatures and total heat source flow rates into the physical information neural network, and outputs the predicted temperature of each heat exchange station after the effect of its own dynamic distribution time lag.

[0067] The predicted temperature of each heat exchange station is compared with the corresponding minimum comfort temperature threshold for each heat exchange station, with the following constraints:

[0068] in, This indicates the predicted temperature for each heat exchange station. This represents a dynamic time delay matrix composed of the dynamic transmission and distribution time delays of all heat exchange stations in the entire network. This indicates the minimum comfort temperature threshold corresponding to each heat exchange station. The minimum comfort temperature threshold is the lowest acceptable secondary network water supply temperature or the calculated lower limit of indoor temperature control for the area under the jurisdiction of each heat exchange station.

[0069] The proposed scheduling schemes that ensure the predicted temperature of all heat exchange stations is not lower than their respective minimum comfort temperature thresholds are selected as candidate scheduling schemes. The scheme that minimizes the objective function value from all candidate scheduling schemes is selected as the optimal scheduling scheme. The heat source supply water temperature and total heat source flow rate corresponding to the optimal scheduling scheme are output as the scheduling instructions for the centralized heating network.

[0070] Preferably, the optimization algorithm is a differential evolution algorithm.

[0071] This invention also provides an intelligent self-organizing production scheduling system for centralized heating networks, such as... Figure 2 As shown, the system includes: a calculation module, a modeling module, and an optimization module.

[0072] The calculation module is used to acquire historical time-series operation data of heat sources and each heat exchange station, calculate the transmission entropy of heat source to each heat exchange station, determine the dynamic transmission and distribution time delay that maximizes the transmission entropy by adjusting the time lag order, obtain the dynamic causal connectivity graph of the whole network, and obtain the dynamic transmission and distribution time delay of each heat exchange station. The modeling module is used to construct a long short-term memory network. It adopts the dynamic transmission and distribution time delay of each heat exchange station, takes the current heat source output data and the historical state data of each heat exchange station under its own dynamic transmission and distribution time delay as input, and takes the predicted temperature field and predicted pressure field of each heat exchange station at the future time as output. It constructs a total loss function including data error loss, continuous mass conservation residual and energy conservation and heat dissipation attenuation physical residual for training, and obtains physical information neural network. The optimization module is used to construct an objective function with the goal of minimizing the sum of the heating energy consumption of the heat source and the power consumption of the main circulation pump. It embeds a physical information neural network as a dynamic implicit constraint function into the optimization algorithm, selects the candidate scheduling scheme that minimizes the objective function value as the optimal scheduling scheme, and outputs the corresponding heat source water supply temperature and total heat source flow rate as scheduling instructions.

[0073] It is understood that the intelligent self-organizing production scheduling system for centralized heating networks provided by the present invention corresponds to the intelligent self-organizing production scheduling method for centralized heating networks provided in the foregoing embodiments. The relevant technical features of the intelligent self-organizing production scheduling system for centralized heating networks can be referred to the relevant technical features of the intelligent self-organizing production scheduling method for centralized heating networks, and will not be repeated here.

[0074] Another object of the present invention is to provide an electronic device, such as... Figure 3 As shown, it includes a memory, a processor, a communication interface, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the intelligent self-organizing production scheduling method for the centralized heating network.

[0075] The intelligent self-organizing production scheduling method for centralized heating networks includes the following steps: Historical time-series operation data of heat sources and each heat exchange station are obtained. The transmission entropy of heat source to each heat exchange station is calculated. The dynamic transmission and distribution time delay that maximizes the transmission entropy is determined by adjusting the time lag order. The dynamic causal connectivity graph of the whole network is obtained, and the dynamic transmission and distribution time delay of each heat exchange station is obtained. A long short-term memory network is constructed, using the dynamic transmission and distribution time delay of each heat exchange station. The current heat source output data and the historical state data of each heat exchange station under their respective dynamic transmission and distribution time delay are used as inputs, and the predicted temperature field and predicted pressure field of each heat exchange station at future time are used as outputs. A total loss function including data error loss, continuous mass conservation residual and energy conservation and heat dissipation attenuation physical residual is constructed for training to obtain a physical information neural network. The objective function is constructed with the goal of minimizing the sum of the heat source heating energy consumption and the main circulation pump power consumption. The physical information neural network is embedded as a dynamic implicit constraint function into the optimization algorithm. The candidate scheduling scheme that minimizes the objective function value is selected as the optimal scheduling scheme, and the corresponding heat source water supply temperature and total heat source flow rate are output as scheduling instructions.

[0076] A fourth objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the intelligent self-organizing production scheduling method for centralized heating networks.

[0077] The intelligent self-organizing production scheduling method for centralized heating networks includes the following steps: Historical time-series operation data of heat sources and each heat exchange station are obtained. The transmission entropy of heat source to each heat exchange station is calculated. The dynamic transmission and distribution time delay that maximizes the transmission entropy is determined by adjusting the time lag order. The dynamic causal connectivity graph of the whole network is obtained, and the dynamic transmission and distribution time delay of each heat exchange station is obtained. A long short-term memory network is constructed, using the dynamic transmission and distribution time delay of each heat exchange station. The current heat source output data and the historical state data of each heat exchange station under their respective dynamic transmission and distribution time delay are used as inputs, and the predicted temperature field and predicted pressure field of each heat exchange station at future time are used as outputs. A total loss function including data error loss, continuous mass conservation residual and energy conservation and heat dissipation attenuation physical residual is constructed for training to obtain a physical information neural network. The objective function is constructed with the goal of minimizing the sum of the heat source heating energy consumption and the main circulation pump power consumption. The physical information neural network is embedded as a dynamic implicit constraint function into the optimization algorithm. The candidate scheduling scheme that minimizes the objective function value is selected as the optimal scheduling scheme, and the corresponding heat source water supply temperature and total heat source flow rate are output as scheduling instructions.

[0078] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0080] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0081] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent self-organizing production scheduling in a centralized heating network, characterized in that, include: Historical time-series operation data of heat sources and each heat exchange station are obtained. The transmission entropy of heat source to each heat exchange station is calculated. The dynamic transmission and distribution time delay that maximizes the transmission entropy is determined by adjusting the time lag order. The dynamic causal connectivity graph of the whole network is obtained, and the dynamic transmission and distribution time delay of each heat exchange station is obtained. A long short-term memory network is constructed, using the dynamic transmission and distribution time delay of each heat exchange station. The current heat source output data and the historical state data of each heat exchange station under their respective dynamic transmission and distribution time delay are used as inputs, and the predicted temperature field and predicted pressure field of each heat exchange station at future time are used as outputs. A total loss function including data error loss, continuous mass conservation residual and energy conservation and heat dissipation attenuation physical residual is constructed for training to obtain a physical information neural network. The objective function is constructed with the goal of minimizing the sum of the heat source heating energy consumption and the main circulation pump power consumption. The physical information neural network is embedded as a dynamic implicit constraint function into the optimization algorithm. The candidate scheduling scheme that minimizes the objective function value is selected as the optimal scheduling scheme, and the corresponding heat source water supply temperature and total heat source flow rate are output as scheduling instructions.

2. The intelligent self-organizing production scheduling method for centralized heating networks according to claim 1, characterized in that, The heat transfer entropy from the heat source to each heat exchange station is: in, Indicates the heat source. Indicates the first A heat exchange station, Indicates heat source to heat exchange station Transmission entropy, Indicates the time of the heat source water supply temperature, Indicates the first Each heat exchange station at time The inlet water temperature, Denotes the joint probability density. and These represent the time lag order, respectively.

3. The intelligent self-organizing production scheduling method for centralized heating networks according to claim 1, characterized in that, The data error loss is: in, Indicates the number of samples. Indicates the actual temperature. Indicates the predicted temperature; The residual for continuous mass conservation is: in, This represents the node index of the heat exchange station. Indicates the inflow of the first Total flow rate of each heat exchange station node Indicates the outflow of the first Total flow rate of each heat exchange station node; The physical residuals of energy conservation and heat dissipation decay are: in, Indicates the node index. Indicates the first Predicted temperature of each node, Indicates time, Indicates the first The flow rate of hot water at each node Indicates the first Predict the spatial gradient of temperature at each node. Indicates the first The heat loss coefficient of the pipe at each node, This indicates the density of hot water. This indicates the specific heat capacity of hot water. This indicates the temperature of the underground soil.

4. The intelligent self-organizing production scheduling method for centralized heating networks according to claim 1, characterized in that, The total loss function is: in, Indicates data error loss. This represents the residual due to continuous mass conservation. This represents the physical residuals due to energy conservation and heat dissipation decay. and They represent and Weighting coefficients in the total loss function.

5. The intelligent self-organizing production scheduling method for centralized heating networks according to claim 1, characterized in that, The objective function is: in, Represents the objective function value. This represents the unit heat cost of natural gas, coal, or industrial waste heat consumed by the heat source. Indicates time Total flow rate of the main circulation pump Indicates time The temperature of the heat source water supply, Indicates time The return water temperature of the heat source, The time-of-use electricity price indicates the amount of electricity consumed by the main circulation pump. Indicates time The main circulation pump pressure differential.

6. The intelligent self-organizing production scheduling method for centralized heating networks according to claim 1, characterized in that, The physical information neural network, acting as a dynamic implicit constraint function, has the following input-output relationship: in, This indicates the predicted temperature for each heat exchange station. This represents a dynamic time delay matrix composed of the dynamic transmission and distribution time delays of all heat exchange stations in the entire network. Represents a neural network of physical information. Indicates time The temperature of the heat source water supply, Indicates time Total flow rate of the main circulation pump.

7. The intelligent self-organizing production scheduling method for centralized heating networks according to claim 1, characterized in that, The process of selecting the optimal scheduling scheme by the optimization algorithm is as follows: Each iteration attempts different heat source water supply temperatures and total heat source flow rates, and the predicted temperature of each heat exchange station is output by the physical information neural network. The schemes that satisfy the condition that the predicted temperature of each heat exchange station is not lower than its own minimum comfort temperature threshold are selected as candidate scheduling schemes, and the candidate scheduling scheme that minimizes the objective function value is selected as the optimal scheduling scheme. The constraint is that the predicted temperature of each heat exchange station is not lower than its respective minimum comfort temperature threshold. in, This indicates the predicted temperature for each heat exchange station. This represents a dynamic time delay matrix composed of the dynamic transmission and distribution time delays of all heat exchange stations in the entire network. This indicates the minimum comfort temperature threshold for each heat exchange station.

8. A centralized heating network intelligent self-organizing production scheduling system, characterized in that, include: The calculation module is used to acquire historical time-series operation data of heat sources and each heat exchange station, calculate the transmission entropy of heat source to each heat exchange station, determine the dynamic transmission and distribution time delay that maximizes the transmission entropy by adjusting the time lag order, obtain the dynamic causal connectivity graph of the whole network, and obtain the dynamic transmission and distribution time delay of each heat exchange station. The modeling module is used to construct a long short-term memory network. It adopts the dynamic transmission and distribution time delay of each heat exchange station, takes the current heat source output data and the historical state data of each heat exchange station under its own dynamic transmission and distribution time delay as input, and takes the predicted temperature field and predicted pressure field of each heat exchange station at the future time as output. It constructs a total loss function including data error loss, continuous mass conservation residual and energy conservation and heat dissipation attenuation physical residual for training, and obtains physical information neural network. The optimization module is used to construct an objective function with the goal of minimizing the sum of the heating energy consumption of the heat source and the power consumption of the main circulation pump. It embeds a physical information neural network as a dynamic implicit constraint function into the optimization algorithm, selects the candidate scheduling scheme that minimizes the objective function value as the optimal scheduling scheme, and outputs the corresponding heat source water supply temperature and total heat source flow rate as scheduling instructions.

9. An electronic device 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 steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images