A method and system for optimizing the operation of multi-heat source networked heating

Abstract

This invention relates to the field of data processing technology, and provides a method and system for optimizing the operation of multi-heat-source networked heating. The method includes: collecting environmental parameters, heat source data, market dynamics, user behavior characteristics, and pipeline topology maps by deploying IoT devices; obtaining a time-series predicted value of heat load for a preset future time period; processing the heat source data and pipeline topology map to obtain a pipeline operation status matrix; aligning the feature dimensions of the pipeline operation status matrix to obtain a pipeline spatial topology feature mapping value; weighting and fusing the heat load time-series predicted value and the pipeline spatial topology feature mapping value to obtain a final heat load predicted value; and optimizing the heating operation based on the final heat load predicted value. This invention enables a real-time response mechanism to environmental parameters, market dynamics, and user behavior characteristics, achieving collaborative optimization of multiple heat sources, improving heat source collaboration efficiency, and reducing energy waste and operating costs.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method and system for optimizing the operation of multi-heat source networked heating. Background Technology

[0002] Single-source heating refers to a heating system in which all heat comes from a single heat source. This heat source provides the heat energy (such as heating, hot water supply, industrial heat, etc.) needed for production and life to users (such as residents, commercial buildings, industrial enterprises, etc.) in a specific area through a pipeline network of heat medium (such as hot water, steam).

[0003] Single-source heating typically distributes heat load in a fixed ratio, lacking a real-time response mechanism to environmental parameters (such as outdoor temperature and humidity), market dynamics (electricity price peaks and valleys, carbon price fluctuations), and user behavior characteristics (room temperature preferences, building thermal inertia), resulting in low heat source synergy efficiency. Summary of the Invention

[0004] Based on this, in order to solve the problem of low heat source coordination efficiency in single heat source heating, the present invention provides a method and system for optimizing the operation of multi-heat source network heating, the specific technical solution of which is as follows:

[0005] A method for optimizing the operation of a multi-heat-source interconnected heating system includes the following steps:

[0006] By deploying IoT devices, environmental parameters, heat source data, market dynamics, user behavior characteristics, and pipeline topology are collected.

[0007] The environmental parameters, market dynamics information, and user behavior characteristics are processed to obtain the predicted heat load time series value for a future preset time period.

[0008] The heat source data and pipeline topology map are processed to obtain the pipeline operation status matrix;

[0009] Based on the dimensional information of the predicted heat load time series values, the feature dimension alignment process is performed on the pipeline network operation status matrix to obtain the pipeline network spatial topology feature mapping value.

[0010] The heat load time series prediction value and the pipeline spatial topology feature mapping value are weighted and fused to obtain the final heat load prediction value;

[0011] Adjust the output value of each heat source according to the final heat load forecast to achieve optimized heating operation.

[0012] The multi-heat-source network heating optimization operation method obtains the final heat load prediction value by performing multi-modal fusion of the heat load time series prediction value and the pipeline network operation status matrix. It can realize a real-time response mechanism to environmental parameters, market dynamics and user behavior characteristics, realize the collaborative optimization of multiple heat sources, improve the collaborative efficiency of heat sources, and reduce energy waste and operating costs.

[0013] Preferably, the specific method for obtaining the predicted heat load time series value for a future preset time period includes the following steps:

[0014] An LSTM model is constructed, and the environmental parameters, market dynamics information, and user behavior characteristics are used as inputs to the LSTM model. The LSTM model outputs the predicted heat load time series values ​​for a future preset time period.

[0015] Preferably, the specific method for obtaining the pipeline network operation status matrix includes the following steps:

[0016] A GCN model is constructed, and the heat source data and pipeline topology map are used as inputs to the GCN model. The pipeline operation status matrix is ​​output through the GCN model.

[0017] Preferably, the specific method for outputting the pipeline network operation status matrix through the GCN model includes the following steps:

[0018] Initialize the node feature matrix and adjacency matrix obtained based on the heat source data and the pipeline topology map;

[0019] After aggregating neighbor node information through graph convolutional layers, multi-layer stacking and feature fusion are performed.

[0020] The node features are normalized to obtain the pipeline network operation status matrix.

[0021] Preferably, the specific method for adjusting the output value of each heat source according to the final heat load forecast to achieve optimized heating operation includes the following steps:

[0022] A fuel cost function is constructed based on the unit fuel cost and fuel consumption, and the economic cost item is obtained based on the fuel cost function.

[0023] A carbon emission cost function is constructed based on the carbon emissions per unit of fuel and the fuel consumption, and the environmental cost item is obtained based on the carbon emission cost function.

[0024] A supply-demand deviation term is constructed based on the final heat load forecast and the target heat supply.

[0025] Based on the economic cost item, environmental cost item, and supply-demand deviation item, a target optimization function is constructed. Based on the target optimization function, the output value of each heat source is adjusted to achieve a balance between the economy, environmental protection, and supply-demand matching of the heating system.

[0026] Preferably, the final heat load prediction L t =λ1·LSTM(T) out ,P grid U habit )+λ2·GCN(G pipe H source )+ε;

[0027] Among them, T out P grid U habit These represent the environmental parameters, market dynamics information, and user behavior characteristics, respectively. LSTM(T) out ,P grid U habit G represents the predicted heat load time series value for a preset future time period. pipe H source These represent the heat source data and the pipeline topology diagram, respectively, GCN (G pipe H source ) represents the mapping value of the spatial topology features of the pipeline network, ε represents the preset model error, and λ1 and λ2 are weight coefficients.

[0028] Preferably, the objective optimization function is expressed as min∑[α·C fuel +β·E emis +γ·||Q tgt -L t ||];

[0029] Among them, C fuel E emis Let Q represent the fuel cost function and the carbon emission cost function, respectively. tgt Target heat supply, α·C fuel β·E emis ,γ·||Q tgt -L t || represent the economic cost item, environmental cost item, and supply-demand deviation item, respectively, and α, β, and γ represent the economic weight coefficient, environmental weight coefficient, and supply-demand balance weight coefficient, respectively.

[0030] A multi-heat-source networked heating optimization operation system, used to implement the aforementioned multi-heat-source networked heating optimization operation method, includes:

[0031] IoT devices are used to collect environmental parameters, heat source data, market dynamics information, user behavior characteristics, and pipeline topology diagrams;

[0032] The heat load prediction module is used to process the environmental parameters, market dynamics information and user behavior characteristics to obtain the heat load time series prediction value for a future preset time period.

[0033] The state matrix acquisition module is used to process the heat source data and the pipeline topology diagram to obtain the pipeline operation state matrix;

[0034] The feature dimension alignment module is used to perform feature dimension alignment processing on the pipeline network operation status matrix based on the dimension information of the heat load time series prediction value, and obtain the pipeline network spatial topology feature mapping value.

[0035] The fusion module is used to perform weighted fusion of the predicted heat load time series value and the pipeline spatial topology feature mapping value to obtain the final predicted heat load value.

[0036] The control module is used to adjust the output value of each heat source according to the final heat load prediction value to achieve optimized operation of the heating supply.

[0037] Preferably, the fusion module is based on formula L t =λ1·LSTM(T) out ,P grid U habit )+λ2·GCN(G pipe H source The final predicted heat load value is obtained by adding ε.

[0038] Where Lt represents the final predicted heat load, and T out P grid U habit These represent the environmental parameters, market dynamics information, and user behavior characteristics, respectively. LSTM(T) out ,P grid U habit G represents the predicted heat load time series value for a preset future time period. pipe H source These represent the heat source data and the pipeline topology diagram, respectively, GCN (G pipe H source ) represents the mapping value of the spatial topology features of the pipeline network, ε represents the preset model error, and λ1 and λ2 are weight coefficients.

[0039] Preferably, the control module includes:

[0040] An economic cost item construction unit is used to construct a fuel cost function based on unit fuel cost and fuel consumption, and to obtain economic cost items based on the fuel cost function.

[0041] An environmental cost item construction unit is used to construct a carbon emission cost function based on the carbon emission per unit of fuel and the fuel consumption, and to obtain environmental cost items based on the carbon emission cost function.

[0042] A supply-demand deviation term construction unit is used to construct a supply-demand deviation term based on the final heat load forecast and the target heat supply.

[0043] The target optimization unit is used to construct a target optimization function based on the economic cost item, the environmental cost item, and the supply and demand deviation item, and adjust the output value of each heat source based on the target optimization function to achieve a balance between the economy, environmental protection, and supply and demand matching of the heating system. Attached Figure Description

[0044] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0045] Figure 1 This is a schematic diagram of the overall process of a multi-heat-source networked heating optimization operation method according to an embodiment of the present invention;

[0046] Figure 2 This is a flowchart illustrating a specific method for outputting a pipeline network operation status matrix through the GCN model in one embodiment of the present invention.

[0047] Figure 3 This is a flowchart illustrating a specific method for optimizing heating operation by adjusting the output value of each heat source according to the final predicted heat load value in one embodiment of the present invention.

[0048] Figure 4 This is a schematic diagram of the overall structure of a multi-heat-source networked heating optimization operation system according to an embodiment of the present invention;

[0049] Figure 5 This is a functional structure diagram of the control module in one embodiment of the present invention;

[0050] Figure 6 This is a functional structure diagram of the state matrix acquisition module in one embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0052] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0054] In this invention, "first" and "second" do not represent a specific quantity or order, but are merely used to distinguish names.

[0055] Before describing the embodiments of the present invention, a brief introduction to the prior art will be given.

[0056] Single-source heating refers to a heating system where all heat comes from a single heat source. This heat source is delivered through a network of heat transfer media (such as hot water or steam) to provide the necessary heat energy (e.g., heating, hot water supply, industrial heat) for production and daily life to users within a specific area (e.g., residents, commercial buildings, industrial enterprises). The single heat source can be a traditional energy source, such as coal / gas-fired boiler rooms, combined heat and power plants (utilizing waste heat from power generation), or oil-fired boilers; or it can be a clean / renewable energy source, such as heat pumps (air source, ground source), solar heating systems, biomass boilers, or industrial waste heat recovery devices.

[0057] A single heat source system has a simple structure. The heat generated by the heat source is generally transported to the heat exchange station or directly supplied to users through a primary pipeline network (main pipeline). The process is simple, and management and control are relatively centralized. It is suitable for small areas, such as small and medium-sized cities, towns, industrial parks or single communities, with relatively small heating areas and high heat source load matching.

[0058] Single heat source heating has the following disadvantages:

[0059] 1. Reliability risk: If the heat source fails (such as equipment maintenance, fuel shortage, or energy supply interruption), the heating of the entire area will be affected, as there is no buffer of backup heat source.

[0060] 2. Limitations of load regulation: The output regulation range of a single heat source is limited, making it difficult to flexibly cope with extreme weather or drastic changes in user load;

[0061] 3. High dependence on pipeline network: Long-distance heat transmission may lead to increased heat loss, and pipeline network failures (such as leaks) will directly affect the stability of heating supply;

[0062] 4. Low efficiency of heat source coordination: Most heat loads are distributed in a fixed ratio, lacking a real-time response mechanism to environmental parameters (such as outdoor temperature and humidity), market dynamics (electricity price peaks and valleys, carbon price fluctuations) and user behavior characteristics (room temperature preferences, building thermal inertia);

[0063] 5. Limited predictive and optimization dimensions: Existing load forecasting is mostly based on historical data statistics (such as the ARIMA model), without integrating network topology characteristics and user behavior data;

[0064] 6. Lack of multi-objective balancing ability: Traditional optimization models focus on economic costs or environmental indicators, and lack multi-objective collaborative optimization of heating network safety (pressure fluctuation threshold) and user comfort (room temperature standard deviation);

[0065] Multi-source network refers to a heating system composed of two or more heat sources that provides heat to users. Its core principle is to allow for the free allocation of heat supply from each heat source on demand, while ensuring the quality of heating for users. Unlike single-source operation and scheduling, the hydraulic and thermal processes involved in multi-source network operation are considerably more complex, significantly increasing the difficulty of operation and scheduling.

[0066] One objective of this invention is to solve the problem of low heat source coordination efficiency in single-heat-source heating, such as... Figure 1 As shown, an embodiment of the present invention provides a method for optimizing the operation of a multi-heat-source networked heating system, comprising the following steps:

[0067] S1 collects environmental parameters, heat source data, market dynamics, user behavior characteristics, and pipeline topology by deploying IoT devices.

[0068] Here, the environmental parameters include, but are not limited to, ambient temperature, ambient humidity, and ambient PM2.5 parameters; the heat source data includes, but is not limited to, heat source (such as boiler, ground source, etc.) efficiency, pipeline pressure and flow, heat source cost, and heat source emission intensity; the market dynamic information includes, but is not limited to, market electricity price and carbon price; the user behavior characteristics include, but are not limited to, user room temperature preference, building structure, insulation level, and user work and rest patterns; and the pipeline topology includes, but is not limited to, nodes (such as heating stations, valves, heat exchangers), edges (such as pipeline connection relationships), and dynamic attributes (such as real-time pressure, temperature, and flow).

[0069] The IoT (Internet of Things) devices mentioned refer to information sensing devices within the Internet of Things. Specifically, ambient temperature and humidity can be collected using temperature and humidity sensors (such as DHT22 / AM2302), which can be distributed and deployed every 50 meters on the exterior walls of buildings. PM2.5 parameters can be collected using laser PM2.5 detectors (such as SDS011) installed at heat source exhaust vents and high points in residential areas.

[0070] Boiler efficiency = (output heat / fuel calorific value) x 100%, requiring simultaneous acquisition of fuel flow rate and steam enthalpy. Boiler efficiency can be acquired by installing a thermal imaging camera and a gas flow meter. Pipeline pressure and flow can be acquired by a piezoelectric pressure transmitter and an electromagnetic flow meter (4-20mA output). User room temperature preferences can be acquired by a smart thermostat using the Nest / Zigbee protocol. Building structure can be obtained by scanning heat loss points in the building with an infrared thermal imager.

[0071] For the aforementioned pipeline topology, BeiDou / GPS dual-mode positioning modules and phase sensors can be installed at heat source stations, pumping stations, and user entrances to collect real-time information such as latitude and longitude coordinates (spatial location), electrical phase (distinguishing three-phase pipeline paths), unique device IDs, and hierarchical relationships (Major / Minor coding). Pressure-temperature composite sensors are deployed every 200m along the pipeline, transmitting pipeline impedance and flow data via NB-IoT. Then, based on node locations and the hydraulic model, a node-branch association matrix is ​​generated, with rows and columns representing nodes and pipelines respectively. Finally, a breadth-first search is used to traverse the association matrix, outputting a weighted network graph, thus obtaining the pipeline topology. When a new device is added, neighboring nodes are automatically scanned, and subnets are displayed in a collapsed manner according to hierarchical rules (root node > heat exchange station > user). If coordinates drift, real-time redrawing is triggered, and device attributes are displayed in a floating manner.

[0072] The heat source data can be defined as a heat source feature matrix H. source ∈R N'×M' Where N' and M' represent the number of heat sources and the characteristic dimension, respectively. The characteristic dimension includes, but is not limited to, thermal efficiency, carbon emission intensity, COP (Coefficient of Performance), and start-stop response time. Thermal efficiency refers to the ratio of fuel input to heat output, carbon emission intensity refers to the CO2 emissions per unit of heat produced, COP refers to the heat output / power consumption of the heat pump, and start-stop response time refers to the delay from the issuance of a command to the equipment operating at full load.

[0073] For sensor parameters collected by IoT devices, data alignment, matrix dimensionality reduction, and feature standardization can be performed. Feature alignment can synchronize sensor data at 5-minute intervals, with missing values ​​compensated using Kalman filtering. Matrix dimensionality reduction involves performing principal component analysis on the original data, retaining feature vectors with a contribution rate >95%. Feature standardization refers to Z-score normalization of each feature dimension to eliminate the influence of dimensions.

[0074] It's important to note that the methods described above are just some of the ways to collect environmental parameters, heat source data, market dynamics, user behavior characteristics, and pipeline topology maps by deploying IoT devices. The methods for deploying IoT devices to collect and process relevant parameter information can be adjusted according to different scenarios or specific implementations, and will not be limited here.

[0075] S2, process the environmental parameters, market dynamics information and user behavior characteristics to obtain the predicted value of heat load time series for a future preset time period.

[0076] As a preferred technical solution, the specific method for obtaining the predicted heat load time series value for a future preset time period includes the following steps: constructing an LSTM model, and using the environmental parameters, market dynamics information, and user behavior characteristics as inputs to the LSTM model, and outputting the predicted heat load time series value for the future preset time period through the LSTM model.

[0077] The number of hidden units in the LSTM model ranges from 64 to 128. The predicted heat load time series output by the LSTM model for a preset future time period can be expressed as LSTM(T) out ,P grid U habit ), T out P grid U habit These respectively represent the environmental parameters, market dynamics information, and user behavior characteristics.

[0078] Specifically, the environmental parameter T out Defined as outdoor temperature time-series data, used to reflect the impact of meteorological conditions on heat demand, it may include information such as historical temperature, temperature change rate, and seasonal characteristics. The market dynamic information P... grid Defined as a time-series signal of grid electricity prices, it affects the economics of heat source operation and may include information such as real-time electricity prices and peak-shaving signals. The user behavior feature U... habit Defined as a user behavior feature matrix, it can include information such as room temperature setting preferences, building thermal inertia parameters, building insulation level, and daily routines.

[0079] The LSTM (Long Short-Term Memory) network extracts the dynamic heat load variation patterns over 24 hours through gating mechanisms (forget gate, input gate, and output gate), and captures the nonlinear impact of events such as sudden temperature drops and peak / valley electricity prices on heating demand. Finally, it outputs the hourly heat load prediction value Q for a preset future time period. pred Here, the hourly heat load forecast value Q for the future preset time period is... pred This can also be understood as a predicted value of heat load time series, with the future preset time period generally set to the next 24 hours. Compared with the traditional ARIMA (Auto Regressive Integrated Moving Average) model, the heat load time series predicted value for the future preset time period output by the LSTM model can handle long-range dependencies in non-stationary time series.

[0080] S3, process the heat source data and pipeline topology map to obtain the pipeline operation status matrix.

[0081] As a preferred technical solution, the specific method for obtaining the pipeline network operation status matrix includes the following steps: constructing a GCN model, and using the heat source data and pipeline network topology map as inputs to the GCN model, and outputting the pipeline network operation status matrix through the GCN model.

[0082] The pipeline network operation state matrix output by the GCN model can be expressed as GCN(G pipe H source )'; where G pipe H source The network topology diagram and heat source data are represented respectively. Specifically, the heat source data can be set in the form of a heat source feature matrix. The columns of the heat source feature matrix are heat source numbers, such as coal-fired units, electric heat pumps, gas-fired units, ground sources, etc., and the columns are heat source features corresponding to different heat source numbers, such as output efficiency, fuel cost, carbon emission intensity, COP coefficient, and start-stop response time.

[0083] Specifically, such as Figure 2 As shown, the specific method for outputting the pipeline network operation status matrix through the GCN model includes the following steps:

[0084] S31. Initialize the node feature matrix and adjacency matrix obtained based on the heat source data and the pipeline topology diagram. Initializing the node feature matrix generally involves normalizing the node feature matrix to eliminate dimensional differences. In some cases, the nodes in the pipeline topology diagram include heat source nodes (such as coal-fired boilers and heat pumps), heat exchange station nodes (such as community heat exchange stations), and user nodes (such as residential buildings). Node features include, but are not limited to, heat source type, output, efficiency, and cost.

[0085] S32, after aggregating neighbor node information through graph convolutional layers, performs multi-layer stacking and feature fusion; here, one of the functions of aggregating neighbor node information through graph convolutional layers is to address the hydraulic balance relationship of the heating network.

[0086] The formula for aggregating neighbor node information in a graph convolutional layer is: in, This is an adjacency matrix with self-joins, used to address the issue of lost node features and ensure that node information is included during aggregation. A and I represent the original adjacency matrix and the identity matrix, respectively. This represents a diagonal matrix used to normalize the adjacency matrix and prevent the propagation of features dominated by nodes with high height. H (l) Let W represent the feature matrix of the nodes in the l-th layer, which serves as the input feature and stores the feature vectors of all nodes in the current layer. (l) Represented as a learnable weight matrix, it is used to perform linear transformations on features and extract higher-order features. σ(·) represents a nonlinear function, typically a ReLU activation function, used to introduce nonlinear expressive power and enhance the model's ability to fit complex patterns.

[0087] Essentially, this involves symmetric normalization of the adjacency matrix, aiming to eliminate the impact of node degree differences on feature scale. For example, nodes with higher elevations aggregate more neighbor information, which requires balancing through the degree matrix.

[0088] Its function is to aggregate neighbor information. The output feature of each node is a weighted average of the features of all its neighbor nodes (including itself). This indicates a nonlinear transformation performed on the aggregated features.

[0089] Multi-layer stacking and feature fusion can be understood as multi-layer GCNs gradually expanding the feature perception range. For example, the first layer of the GCN captures the influence of directly connected nodes (such as heat source → user), the second layer of the GCN captures the influence of indirect nodes (such as heat source → user → adjacent building), and the final layer outputs a d-dimensional vector G. i ∈R d .

[0090] Generally, a 3-layer GCN model structure uses the first layer as the input layer, the second layer as an intermediate hidden layer, and the third layer as the output layer. Since the specific structure of a 3-layer GCN model is standard technology in this field, it will not be elaborated upon here.

[0091] S33, normalize the node features to obtain the pipeline network operation state matrix. The final pipeline network operation state matrix is ​​expressed as [G1, G2, ..., G...]. nHere, n represents the number of nodes in the pipeline network. The normalization method can be node-by-node L2 normalization: This ensures that the feature vectors of different nodes are comparable.

[0092] S4. Based on the dimensional information of the predicted heat load time series values, the feature dimension alignment process is performed on the pipeline network operation status matrix to obtain the pipeline network spatial topology feature mapping value.

[0093] In many cases, the predicted heat load time series values ​​do not match the dimensions of the pipeline network operation status matrix. To perform weighted fusion, feature dimension alignment is required. Methods for feature dimension alignment include, but are not limited to, fully connected layer mapping and pooling operations.

[0094] For the fully connected layer mapping method: Fully connected layers can be added after the output layers of LSTM and GCN, mapping them to a unified dimension. For example, if the output dimension of LSTM is d... lstm The output dimension of GCN is d gcn It can be uniformly mapped to the same dimension d through the FC layer (fully connected layer). common For pooling operations: global average pooling or max pooling is performed on the high-dimensional output to compress it into a low-dimensional vector. For example, the output of GCN is an N×d feature matrix of graph nodes. gcn 1×d is obtained through pooling. gcn The global features are then mapped to match the output dimension of the LSTM.

[0095] In this example, the pipeline network operation state matrix can be pooled first. By performing global average pooling or max pooling on the high-dimensional output, the pipeline network operation state matrix can be compressed into a low-dimensional spatial feature vector. Then, based on the dimensional information of the heat load time series prediction value, the low-dimensional spatial feature vector can be mapped to the pipeline network spatial topology feature mapping value with the same dimension as the heat load time series prediction value through a fully connected layer.

[0096] S5, the predicted heat load time series value and the pipeline spatial topology feature mapping value are weighted and fused to obtain the final predicted heat load value.

[0097] As a preferred technical solution, the final predicted heat load value is based on formula L. t =λ1·LSTM(T) out ,P grid U habit )+λ2·GCN(G pipe H sourceIt is obtained by calculation of )+ε.

[0098] Among them, LSTM(T) out ,P grid U habit GCN represents the predicted heat load time series value for a preset future time period. pipe H source ) represents the spatial topology feature mapping value of the pipeline network, ε represents the preset model error, and L t λ1 and λ2 represent the final predicted heat load value, respectively, and are the weighting coefficients of the predicted heat load time series value and the pipeline spatial topology feature mapping value for a future preset time period.

[0099] The model error includes disturbances not explicitly considered in the model, including but not limited to sudden heat loss (such as pipeline leakage), extreme weather events (cold waves, blizzards), and sudden changes in user behavior (holiday mode switching).

[0100] The model error ε~N(0,σ) 2 ), indicating that it follows a pattern with a mean of 0 and a variance of σ. 2 The normal distribution is used to characterize stochastic factors not modeled in heat load prediction models, such as sudden equipment failures, meteorological changes, or data measurement errors. The assumption of a normal distribution ensures the validity of statistical inferences (e.g., confidence interval calculations). This represents the standard deviation of the error term, indicating the relationship between the standard deviation of the error term and the historical load mean. Proportional rather than a fixed value. This setting allows the error scale to vary with the load level, avoiding problems with absolute values ​​that are too small or too large. For example, when hour, It corresponds to the relative proportion of actual load fluctuations.

[0101] Here, the coefficient 0.05 in the standard deviation of the error term is determined through fitting historical data, ensuring that the error term covers approximately 95% of the load fluctuation range (based on the 3σ principle of normal distribution). The model error ε ~ N(0,σ) 2 The model can be dynamically updated through an online learning mechanism, and model retraining is triggered when the absolute value of the error exceeds 2σ.

[0102] For example, a simple weighted summation can be performed on the predicted heat load time series value and the pipeline spatial topology feature mapping value. Assuming the weight coefficients λ1 and λ2 are set to 0.75 and 0.25 respectively based on domain experience, and t is 8:00 AM (e.g., morning rush hour in an office area), the predicted heat load value output by the LSTM model is 1100MW, the pipeline spatial topology feature mapping value output by the GCN model is 1050MW, and the model error is 10MW, then the final predicted heat load value at this time point is 0.75×1100 + 0.25×1050 + 5 = 1097.5MW. This method of obtaining the final predicted heat load value through simple weighted summation is computationally simple and easy to interpret, and is suitable for scenarios involving preliminary exploration and clear feature contributions.

[0103] The weighting coefficients λ1 and λ2 can also be dynamically adjusted based on the attention mechanism, assigning dynamic weights (which change with the characteristics of each moment) to the LSTM and GCN outputs at each time point, focusing on the modes that have a significant impact on the current heat load. For example, during the evening peak (6 PM), user habits have a significant impact, so the attention weights are tilted towards the LSTM, with λ1 and λ2 set to 0.8 and 0.2 respectively. At 3 AM, the heat storage in the pipeline network has a significant impact, so the attention weights are tilted towards the GCN, with λ1 and λ2 set to 0.3 and 0.7 respectively.

[0104] At 6 PM, during the evening peak (e.g., the evening peak for residential heating), the network output heat load prediction is 1500MW, and the GCN model outputs a network spatial topology feature mapping value of 1400MW. Therefore, the corresponding final heat load prediction value is 0.8×1500 + 0.2×1400 + 10 = 1490MW. At 3 AM, the LSTM model outputs a heat load prediction of 500MW, and the GCN model outputs a network spatial topology feature mapping value of 550MW. Therefore, the corresponding final heat load prediction value at this time is 0.3×500 + 0.7×550 + 10 = 545MW.

[0105] This method, which uses an attention mechanism to dynamically adjust weight coefficients to obtain the final heat load prediction, can dynamically adjust weights, focus on important features, and make the weight allocation more flexible in scenarios where the contribution of these features changes over time (such as peak / valley).

[0106] S6, adjust the output value of each heat source according to the final heat load prediction value to achieve optimized operation of heating.

[0107] As a preferred technical solution, such as Figure 3 As shown, the specific method for optimizing heating operation by adjusting the output values ​​of each heat source based on the final predicted heat load includes the following steps:

[0108] S61, construct a fuel cost function based on unit fuel cost and fuel consumption, and obtain the economic cost item based on the fuel cost function.

[0109] The fuel cost function is positively correlated with the energy consumption of the heat source type (coal, gas, ground source heat pump, etc.). Assume the unit fuel cost of the i-th heat source is c. i Then the fuel cost function C fuel =∑(c i ·Q i ), where Q i The fuel consumption of the i-th heat source is represented by the economic weighting coefficient α, which can be dynamically adjusted according to market electricity prices and fuel prices (e.g., the value of α is reduced during periods of low electricity prices to encourage electricity consumption).

[0110] S62, construct a carbon emission cost function based on the carbon emissions per unit of fuel and the fuel consumption, and obtain the environmental protection cost item based on the carbon emission cost function.

[0111] The carbon emission cost function is related to the carbon emission intensity of the heat source. Assume EF i The carbon emission cost function E is the carbon emission per unit of fuel relative to the i-th heat source. emis =p emis ∑(EF i ·Q i ), where p emis This represents the average price in the carbon market. The environmental weighting coefficient β changes dynamically with carbon price policies or environmental regulations. For example, when carbon prices rise, β can be increased to prioritize reducing carbon emissions.

[0112] S63, construct a supply-demand deviation term based on the final predicted heat load and the target heat supply.

[0113] S64. Construct a target optimization function based on the economic cost item, environmental cost item, and supply-demand deviation item. Adjust the output value of each heat source based on the target optimization function to achieve a balance between the economy, environmental protection, and supply-demand matching of the heating system.

[0114] Preferably, the objective optimization function is expressed as min∑[α·C fuel +β·E emis +γ·||Q tgt -L t ||];wherein, C fuel E emis Let Q represent the fuel cost function and the carbon emission cost function, respectively. tgt α·C represents the target heat supply. fuel β·E emis ,γ·||Q tgt -L t|| represents the economic cost item, environmental cost item, and supply-demand deviation item, respectively; α, β, and γ represent the economic weight coefficient, environmental weight coefficient, and supply-demand balance weight coefficient, respectively; and ||·|| represents the norm, typically L1 or L2 norm. The supply-demand balance weight coefficient is used to penalize prediction bias and avoid insufficient heating or energy waste.

[0115] The target heat supply can be understood as the actual heat supply expected to be delivered to the target object. For example, if the final predicted heat load of the target object (such as an office area or a residential area) at a certain point in time is 150MW, and based on this final predicted heat load, iterative calculations are performed using the target optimization function to adjust the output of multiple heat sources. The result is that when 145MW is actually delivered to the target object, the overall cost (i.e., the target optimization function value) is minimized. The 145MW actually delivered to the target object can be understood as the target heat supply.

[0116] Among them, Q tgt_i L t_i These represent the target heat supply and the final predicted heat load for the i-th time period, respectively, where m represents the number of future preset time periods. For example, when m = 24, the final predicted heat load L... t This includes the heat load forecast for the next 24 hours, i.e., the heat load forecast for the next 24 time points. d is the preset supply-demand deviation loss, which can be set based on user complaint costs or on industry experience, for example, set to 0.5 yuan / MW·h.

[0117] As a preferred technical solution, the optimized operation of the multi-heat-source network heating system also includes setting heat source output constraints. Where, σ safe This represents the safety redundancy factor, typically set between 0.1 and 0.2, used to address the risks of extreme weather or equipment failure and ensure heating reliability. ret This indicates the return water temperature of the pipeline network, which affects the system's thermal efficiency and equipment lifespan. The lower limit of the return water temperature is set at 45 degrees Celsius to prevent low-temperature corrosion and frost formation, while the upper limit is set at 65 degrees Celsius, primarily to avoid pump cavitation and excessive heat loss from the pipeline network. The upper and lower limits of the aforementioned return water temperature range can be adjusted appropriately based on actual conditions, which will not be elaborated upon here.

[0118] ΔP pipe ≤ρ max Indicates the constraint on pipeline pressure fluctuation, ΔP pipe This represents the pressure difference between nodes in the pipeline network and can be calculated in real time using a fluid dynamics model. ρ max This indicates the maximum permissible differential pressure threshold, and its main function is to prevent hydraulic imbalance or pipeline rupture. Generally, it is set to 120% of the design differential pressure.

[0119] The objective optimization function, by setting economic weight coefficients, environmental weight coefficients, and supply-demand balance weight coefficients, allows α, β, and γ to be adjusted in real time according to external parameters (such as carbon price and fuel price), reflecting multi-objective game optimization. By setting a safety redundancy coefficient and a maximum allowable pressure difference threshold, it possesses a dual-guarantee safety redundancy mechanism for the stability of the heating network, adapting to complex operating conditions involving multiple heat sources interconnected.

[0120] The multi-heat-source network heating optimization operation method obtains the final heat load prediction value by performing multi-modal fusion of the heat load time series prediction value and the pipeline network operation status matrix. It can realize a real-time response mechanism to environmental parameters, market dynamics and user behavior characteristics, realize the collaborative optimization of multiple heat sources, improve the collaborative efficiency of heat sources, and reduce energy waste and operating costs.

[0121] As a preferred technical solution, the economic weight coefficient α = softmax(W[Q) pred ||S pipe ]); where S pipe This represents the low-dimensional feature vector obtained after pooling the pipeline network operation state matrix, containing node pressure, flow rate, and topology constraint parameters. W represents the learnable parameter matrix, used to map the concatenated features to the attention weight space. || represents the feature concatenation operation, used to merge Q along the feature dimension. pred and S pipe This allows for high-dimensional representation. The economic weight coefficients here are normalized attention weight vectors, which can be used to represent the relative importance of the two types of features.

[0122] Specifically, first Q pred and S pipe The concatenation is performed along the feature dimension to form a joint feature vector. Then, the concatenated features are projected into the latent space through a learnable matrix W. Finally, a softmax operation is performed on the linear transformation result to generate attention weights and obtain the final economic weight coefficients.

[0123] The formula α = softmax(W[Q) is used. pred ||S pipe The economic weighting coefficient α changes dynamically with input characteristics, for example, increasing Q during extreme weather. pred Weighting, with emphasis on S during pipeline network failures. pipe It breaks through the limitations of traditional single-modal models and captures the temporal-spatial coupling relationship through an explicit attention mechanism. Compared with multi-head attention mechanisms, the computational complexity of single splicing + linear transformation is lower, making it suitable for real-time heating scheduling scenarios.

[0124] As a preferred technical solution, the multi-heat source network heating optimization operation method further includes the following steps: Pareto front search based on the NSGA-III algorithm to optimize the economic weight coefficient α, so that the economic weight coefficient is adjusted in real time with electricity price and carbon price, thereby achieving adaptive balance between economic and environmental protection objectives and multi-objective dynamic game optimization.

[0125] Specifically β = 1 - α - γ; where C t This represents economic costs, including fuel costs (coal / natural gas), equipment operation and maintenance costs, and electricity purchase costs, which are affected by real-time electricity prices. When electricity prices rise, C... t Significantly increased. E t This represents environmental cost, used to quantify the environmental damage caused by carbon emissions or other pollutants. Environmental cost here can be expressed by the formula E. t =Calculated by carbon emissions x real-time carbon price. The higher the carbon price, the higher the E. t The greater the impact on the objective function, the better. 0.1 and 0.05 represent exponential coefficients, and e represents the natural constant. This represents the economic cost sensitivity adjustment factor; the larger the value, the more sensitive it is to cost changes. This represents the environmental cost sensitivity adjustment factor. A smaller value indicates that environmental goals require greater cost fluctuations to trigger weight adjustments.

[0126] Here, the formula The purpose of the exponential function is explained below: Exponential function as well as This is used to convert costs and emissions into normalized utility values, avoiding optimization biases caused by differences in dimensions, and also avoiding scheduling oscillations caused by fluctuations in market parameters. The exponential coefficient 0.1 in the middle is greater than The exponential coefficient of 0.05 reflects a decision-making tendency that prioritizes economic goals. The exponential coefficients of both can be preset based on experience or obtained through training with historical data.

[0127] function It has an adaptive weighting mechanism; when electricity prices surge, C... t Rapidly increase → Rapid decay leads to a decrease in α, and the system automatically lowers the priority of economic objectives to avoid high-cost operation. When carbon prices rise, E... t Significant increase → As the decay rate increases (β), the system prioritizes emission reduction. The Pareto front search involves real-time weight adjustments to guide the NSGA-III algorithm in searching different regions of the target space: focusing on low-cost solutions (α↑) during periods of high electricity prices, and focusing on low-carbon solutions (β↑) during periods of high carbon prices.

[0128] During peak electricity price periods, when C tWhen α increases and then decreases, energy storage or ground-source heat pumps can be prioritized to reduce the need for expensive electricity purchases; when the carbon trading market is active, E... t When β decreases, the output of coal-fired units can be reduced, and the proportion of clean energy can be increased.

[0129] As a preferred technical solution, the preset model error is calculated according to the following formula:

[0130]

[0131] Where, α'=σ(LSTM(T) out )⊙GCN(G pipe LSTM(T) out This indicates that outdoor temperature time-series data is processed using LSTM to capture disturbances related to sudden temperature drops. GCN(G) pipe This indicates that a graph convolutional network (GCN) is used to process the pipeline topology graph, model the pipeline topology structure, and identify key nodes (such as the impact of valve leakage on downstream nodes). LSTM(T) out )⊙GCN(G pipe The ) indicates that by fusing spatiotemporal features through element-wise multiplication ⊙, the Sigmoid function is normalized to the [0,1] interval. If LSTM(T) out ) and GCN (G pipe The dimensions of LSTM(T) do not match, so first apply the LSTM(T) to the LSTM(T) out ) and GCN (G pipe Feature dimension alignment is performed, followed by LSTM(T) processing. out ) and GCN (G pipe ) element-wise multiplication between ).

[0132] This represents the error term, which is the difference between the actual observed value and the model prediction value. It reflects the real-time deviation of parameters such as temperature and flow rate in the heating network, including explainable physical disturbances (such as thermal inertia delay) and random noise (such as sensor drift).

[0133] loss function Where δ represents the parameter threshold, which is usually set to a value range of 0.1-1.0.

[0134] Quantile loss function Among them, the ability to capture tail risks by quantile τ, such as when τ = 0.9, focuses on extreme errors above the 90th quantile.

[0135] Under normal operating conditions (such as during a stable heating season), α'≈0.7, with Huber loss dominating; in extreme events (such as a sudden cold wave), α' dynamically decreases to 0.3, and the weight of quantile loss increases.

[0136] Huber loss (smoothed L1 loss) exhibits mean squared error (MSE) under small error conditions and is robust to outliers. Quantile loss, on the other hand, captures tail risk through the quantile τ (e.g., τ = 0.9) and is suitable for low-probability, high-impact events such as extreme weather. Therefore, a balance between robustness and sensitivity can be achieved through both Huber and quantile loss functions. The weight α' is dynamically adjusted through a spatiotemporally coupled network, allowing the model to emphasize Huber loss under normal operating conditions and enhance the contribution of quantile loss in the event of sudden failures or extreme events.

[0137] formula By dynamically fusing Huber loss and quantile loss, deterministic errors (such as temperature prediction deviations) and stochastic errors (such as sudden disturbances caused by pipeline leaks) of heating network parameters can be jointly optimized, achieving differentiated modeling of conventional errors and extreme event errors.

[0138] like Figure 4 As shown, an embodiment of the present invention provides a multi-heat source networked heating optimization operation system for implementing the multi-heat source networked heating optimization operation method. The system includes an IoT device, a heat load prediction module, a state matrix acquisition module, a feature dimension alignment module, a fusion module, and a control module.

[0139] IoT devices are used to collect environmental parameters, heat source data, market dynamics, user behavior characteristics, and pipeline topology maps. The heat load prediction module processes the environmental parameters, market dynamics, and user behavior characteristics to obtain the predicted heat load time series values ​​for a preset future time period.

[0140] Specifically, the predicted heat load time series value Q for a future preset time period pred =LSTM(T) out ,P grid U habit ), where T out P grid U habit These respectively represent the environmental parameters, market dynamics information, and user behavior characteristics.

[0141] Assume the input data for the LSTM is as shown in the table below:

[0142] time <![CDATA[T out (℃)]]> <![CDATA[P grid (kWh / kWh) <![CDATA[U habit (Set temperature ℃) t+1 -5 0.48 27 t+2 -4 0.52 26 ... ... ... ... t+24 3 0.65 25

[0143] The calculation process for obtaining the predicted heat load time series value for a future preset time period by processing the environmental parameters, market dynamics, and user behavior characteristics using an LSTM model is as follows: The LSTM model encodes the 24-hour series, stores long-term dependencies (such as the lag effect of heating demand caused by low nighttime temperatures) in memory units, and finally outputs a 24-hour predicted heat load vector, such as Q. pred =LSTM(T) out ,P grid U habit )=[Q t+1 Q t+1 ,…,Q t+24 ]=[1200,1150,…,980](MW).

[0144] The state matrix acquisition module is used to process the heat source data and the pipeline topology map to obtain the pipeline operation state matrix.

[0145] Specifically, such as Figure 6 As shown, the state matrix acquisition module includes an initialization unit, a graph convolutional layer unit, and a normalization unit. The initialization unit is used to initialize the node feature matrix and adjacency matrix obtained based on the heat source data and the pipeline topology map. Initializing the node feature matrix generally involves normalizing the node feature matrix to eliminate dimensional differences.

[0146] For the node feature matrix, if it is a heat source node, it can be expressed as [heat source ID, supply water temperature, flow rate], and if it is a user node, it can be expressed as [user ID, return water temperature, building thermal resistance coefficient].

[0147] Graph convolutional layers are used to perform multi-layer stacking and feature fusion after aggregating neighbor node information; here, one of the purposes of aggregating neighbor node information through graph convolutional layers is to address the hydraulic balance relationship in a thermal network. The formula for aggregating neighbor node information through graph convolutional layers is as follows: in, This is an adjacency matrix with self-joins, used to address the issue of lost node features and ensure that node information is included during aggregation. A and I represent the original adjacency matrix and the identity matrix, respectively. This represents a diagonal matrix used to normalize the adjacency matrix and prevent the propagation of features dominated by nodes with high elevation.

[0148] H (l) Let W represent the feature matrix of the nodes in the l-th layer, which serves as the input feature and stores the feature vectors of all nodes in the current layer. (l) Represented as a learnable weight matrix, it is used to perform linear transformations on features and extract higher-order features. σ(·) represents a nonlinear function, typically a ReLU activation function, used to introduce nonlinear expressive power and enhance the model's ability to fit complex patterns.

[0149] Essentially, this involves symmetric normalization of the adjacency matrix, aiming to eliminate the impact of node degree differences on feature scale. For example, nodes with higher elevations aggregate more neighbor information, which requires balancing through the degree matrix. Its function is to aggregate neighbor information. The output feature of each node is a weighted average of the features of all its neighbor nodes (including itself). This indicates a nonlinear transformation performed on the aggregated features.

[0150] The normalization unit is used to normalize the node features to obtain the pipeline network operation state matrix. The final pipeline network operation state matrix is ​​expressed as [G1, G2, ..., G...]. n ].

[0151] The feature dimension alignment module is used to perform feature dimension alignment processing on the pipeline network operation status matrix based on the dimension information of the heat load time series prediction value, and obtain the pipeline network spatial topology feature mapping value.

[0152] The feature dimension alignment module includes a pooling layer and a fully connected layer. The pooling layer is used to perform pooling operations on the pipeline network operation state matrix. By performing global average pooling or max pooling on the high-dimensional output, the pipeline network operation state matrix is ​​compressed into a low-dimensional spatial feature vector. The fully connected layer is used to map the low-dimensional spatial feature vector to a pipeline network spatial topology feature mapping value with the same dimension as the heat load time series prediction value, based on the dimensional information of the heat load time series prediction value.

[0153] The fusion module is used to perform weighted fusion of the predicted heat load time series value and the pipeline spatial topology feature mapping value to obtain the final predicted heat load value.

[0154] The fusion module is based on formula L t =λ1·LSTM(T) out ,P grid U habit )+λ2·GCN(G pipe H source The final predicted heat load value is obtained by adding ε.

[0155] Where Lt represents the final predicted heat load, and T out P grid U habit These represent the environmental parameters, market dynamics information, and user behavior characteristics, respectively. LSTM(T) out ,P grid U habit G represents the predicted heat load time series value for a future preset time period after feature dimension alignment processing.pipe H source These represent the heat source data and the pipeline topology diagram, respectively, GCN (G pipe H source ) represents the mapping value of the spatial topology features of the pipeline network, ε represents the preset model error, and λ1 and λ2 are weight coefficients.

[0156] The control module is used to adjust the output value of each heat source according to the final predicted heat load to achieve optimized heating operation. Specifically, such as... Figure 5 As shown, the control module includes an economic cost item construction unit, an environmental cost item construction unit, a supply and demand deviation item construction unit, and a target optimization unit.

[0157] The economic cost item construction unit is used to construct a fuel cost function based on the unit fuel cost and fuel consumption, and obtain the economic cost item based on the fuel cost function; the environmental cost item construction unit is used to construct a carbon emission cost function based on the carbon emissions per unit fuel and fuel consumption, and obtain the environmental cost item based on the carbon emission cost function.

[0158] The supply-demand deviation term construction unit is used to construct a supply-demand deviation term based on the final heat load forecast value and the target heat supply; the target optimization unit is used to construct a target optimization function based on the economic cost term, environmental cost term, and supply-demand deviation term, and adjust the output value of each heat source based on the target optimization function to achieve a balance between the economy, environmental protection, and supply-demand matching of the heating system.

[0159] The objective function is expressed as min∑[α·C fuel +β·E emis +γ·||Q tgt -L t ||];wherein, C fuel E emis Let Q represent the fuel cost function and the carbon emission cost function, respectively. tgt Q represents the target heat supply. pred α·C represents the predicted value of heat load time series. fuel β·E emis ,γ·||Q tgt -L t || represent the economic cost item, environmental cost item, and supply-demand deviation item, respectively, while α, β, and γ represent the economic weight coefficient, environmental weight coefficient, and supply-demand balance weight coefficient, respectively. The supply-demand balance weight coefficient is used to penalize forecast deviations and avoid insufficient heating or energy waste.

[0160] For example, suppose the cost of a unit of fuel is c. i Then the fuel cost function C fuel =∑(c i ·Qi ), where Q i The fuel consumption of the i-th heat source is represented by the economic weighting coefficient α, which can be dynamically adjusted according to market electricity prices and fuel prices (e.g., the value of α is reduced during periods of low electricity prices to encourage electricity consumption).

[0161] Assume EF i The carbon emission cost function E is the carbon emission per unit of fuel relative to the i-th heat source. emis =∑(EF) i ·Q i The environmental weighting coefficient β changes dynamically with carbon price policies or environmental regulations. For example, when carbon prices rise, β can be increased to prioritize reducing carbon emissions.

[0162] For example, suppose the system includes three heat sources, and its basic parameters are shown in the table below:

[0163]

[0164] Assume the final predicted heat load for the target object is L at 1 AM, 12 PM, 8 PM, and other off-peak periods. t_1 =1000MW, L t_12 =2000MW, L t_20 =1800MW and L t_其它 =1500MW, target heating capacity Q at 1:00 AM tgt_1 Assuming a total capacity of 1100MW, with the output of three heat sources—coal-fired power plant, boiler room, and ground source heat pump—distributed to 800MW, 200MW, and 100MW respectively, and the average carbon market price at 0.2 yuan / ton CO2, and weighting coefficients α, β, and γ set at 0.4, 0.3, and 0.3 respectively, and a unit deviation loss d = 0.5 yuan / MW·h, then the fuel cost function C fuel =∑(c i ·Q i The carbon emission cost function E is calculated as follows: E = 0.15 × 800 + 0.3 × 200 + 0.2 × 100 = 200 yuan. emis =p emis ∑(EF i ·Q i = 0.2 × (0.8 × 800 + 0.5 × 200 + 0.1 × 100) = 150 yuan, supply and demand deviation cost d·|Q tgt_1 -L t_1 |=0.5×|1100-1000|=50.

[0165] Since the comprehensive cost consists of economic cost, environmental cost, and supply-demand imbalance, the objective function α·C fuel +β·E emis +γ·||Q tgt -Qpred ||=0.4×200+0.3×150+0.3×50=140 yuan.

[0166] The target heat supply Q at noon tgt_12 Assuming a total capacity of 2200MW, with the output of three heat sources—coal-fired power plant, boiler room, and ground source heat pump—distributed to 1500MW, 500MW, and 200MW respectively, and the average carbon market price at 0.2 yuan / ton CO2, and weighting coefficients α, β, and γ set at 0.4, 0.3, and 0.3 respectively, and a unit deviation loss d = 0.5 yuan / MW·h, then the fuel cost function C fuel =∑(c i ·Q i The carbon emission cost function E = 0.15 × 1500 + 0.3 × 500 + 0.2 × 200 = 415 yuan. emis =p emis ∑(EF i ·Q i = 0.2 × (0.8 × 1500 + 0.5 × 500 + 0.1 × 200) = 294 yuan, supply and demand deviation cost d·|Q tgt_1 -L t_1 |=0.5×|2200-2000|=100.

[0167] Since the comprehensive cost consists of economic cost, environmental cost, and supply-demand imbalance, the objective function α·C fuel +β·E emis +γ·||Q tgt -Q pred ||=0.4×415+0.3×294+0.3×100=284.2 yuan.

[0168] By calculating the objective function value at each time point and then summing up the multiple objective function values, the comprehensive objective function value ∑[α·C] can be obtained. fuel +β·E emis +γ·||Q tgt -L t Based on the objective optimization function, the optimal output value of each heat source can be obtained by solving it through continuous iterative optimization. This allows for cost reduction, carbon emission control, and ensuring supply and demand matching for the target object.

[0169] As a preferred technical solution, the multi-heat-source networked heating optimization operation system in this embodiment also includes a heat source output constraint model.

[0170] The heat source output constraint model is expressed as follows: Where, σ safeThis represents the safety redundancy factor, typically set between 0.1 and 0.2, used to address the risks of extreme weather or equipment failure and ensure heating reliability. ret This indicates the return water temperature of the pipeline network, which affects the system's thermal efficiency and equipment lifespan. The lower limit of the return water temperature is set at 45 degrees Celsius to prevent low-temperature corrosion and frost formation, while the upper limit is set at 65 degrees Celsius, primarily to avoid pump cavitation and excessive heat loss from the pipeline network. The upper and lower limits of the aforementioned return water temperature range can be adjusted appropriately based on actual conditions, which will not be elaborated upon here.

[0171] ΔP pipe ≤ρ max Indicates the constraint on pipeline pressure fluctuation, ΔP pipe This represents the pressure difference between nodes in the pipeline network and can be calculated in real time using a fluid dynamics model. ρ max This indicates the maximum permissible differential pressure threshold, and its main function is to prevent hydraulic imbalance or pipeline rupture. Generally, it is set to 120% of the design differential pressure.

[0172] In summary, the multi-heat-source networked heating optimization operation system obtains the final heat load prediction value by performing multimodal fusion of the heat load time series prediction value and the pipeline network operation status matrix. It can realize a real-time response mechanism to environmental parameters, market dynamics and user behavior characteristics, and can achieve collaborative optimization of multiple heat sources, improve heat source collaboration efficiency, and reduce energy waste and operating costs.

[0173] The technical features of the embodiments described can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0174] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for optimizing the operation of a multi-heat-source networked heating system, characterized in that, Includes the following steps: By deploying IoT devices, environmental parameters, heat source data, market dynamics, user behavior characteristics, and pipeline topology are collected. The environmental parameters, market dynamics information, and user behavior characteristics are processed to obtain the predicted heat load time series value for a future preset time period. The heat source data and pipeline topology map are processed to obtain the pipeline operation status matrix; Based on the dimensional information of the predicted heat load time series values, the feature dimension alignment process is performed on the pipeline network operation status matrix to obtain the pipeline network spatial topology feature mapping value. The heat load time series prediction value and the pipeline spatial topology feature mapping value are weighted and fused to obtain the final heat load prediction value; Adjust the output value of each heat source according to the final heat load forecast to achieve optimized heating operation; The specific method for obtaining the predicted heat load time series values ​​for a future preset time period includes the following steps: An LSTM model is constructed, and the environmental parameters, market dynamics information, and user behavior characteristics are used as inputs to the LSTM model. The LSTM model outputs the predicted heat load time series values ​​for a future preset time period. The specific method for obtaining the pipeline network operation status matrix includes the following steps: A GCN model is constructed, and the heat source data and pipeline topology map are used as inputs to the GCN model. The pipeline operation status matrix is ​​output through the GCN model. The specific method for optimizing heating operation by adjusting the output of each heat source based on the final heat load forecast includes the following steps: A fuel cost function is constructed based on the unit fuel cost and fuel consumption, and the economic cost item is obtained based on the fuel cost function. A carbon emission cost function is constructed based on the carbon emissions per unit of fuel and the fuel consumption, and the environmental cost item is obtained based on the carbon emission cost function. A supply-demand deviation term is constructed based on the final heat load forecast and the target heat supply. Based on the economic cost item, environmental cost item, and supply-demand deviation item, a target optimization function is constructed. Based on the target optimization function, the output value of each heat source is adjusted to achieve a balance between the economy, environmental protection, and supply-demand matching of the heating system.

2. The method for optimizing the operation of a multi-heat source networked heating system as described in claim 1, characterized in that, The specific method for outputting the pipeline network operation status matrix through the GCN model includes the following steps: Initialize the node feature matrix and adjacency matrix obtained based on the heat source data and the pipeline topology map; After aggregating neighbor node information through graph convolutional layers, multi-layer stacking and feature fusion are performed. The node features are normalized to obtain the pipeline network operation status matrix.

3. The method for optimizing the operation of a multi-heat-source networked heating system as described in claim 2, characterized in that, Final heat load forecast ; in, These respectively represent the environmental parameters, market dynamics information, and user behavior characteristics. This represents the predicted heat load time series value for a predetermined future time period. These represent the heat source data and the pipeline topology diagram, respectively. Represents the spatial topology feature mapping value of the pipeline network. This represents the preset model error. These are the weighting coefficients.

4. The method for optimizing the operation of a multi-heat source networked heating system as described in claim 3, characterized in that, The objective optimization function is expressed as: ; in, Let these represent the fuel cost function and the carbon emission cost function, respectively. Indicates the target heat supply. These respectively represent the economic cost item, the environmental cost item, and the supply-demand deviation item. These represent the economic weighting coefficient, the environmental weighting coefficient, and the supply-demand balance weighting coefficient, respectively.

5. A multi-heat-source networked heating optimization operation system, used to implement the multi-heat-source networked heating optimization operation method as described in any one of claims 1-4, characterized in that, include: IoT devices are used to collect environmental parameters, heat source data, market dynamics information, user behavior characteristics, and pipeline topology diagrams; The heat load prediction module is used to process the environmental parameters, market dynamics information and user behavior characteristics to obtain the heat load time series prediction value for a future preset time period. The state matrix acquisition module is used to process the heat source data and the pipeline topology diagram to obtain the pipeline operation state matrix; The feature dimension alignment module is used to perform feature dimension alignment processing on the pipeline network operation status matrix based on the dimension information of the heat load time series prediction value, and obtain the pipeline network spatial topology feature mapping value. The fusion module is used to perform weighted fusion of the predicted heat load time series value and the pipeline spatial topology feature mapping value to obtain the final predicted heat load value. The control module is used to adjust the output value of each heat source according to the final heat load prediction value to achieve optimized operation of the heating supply.

6. The multi-heat source networked heating optimized operation system as described in claim 5, characterized in that, The fusion module is based on the formula Obtain the final heat load forecast; in, This represents the final predicted heat load value. These respectively represent the environmental parameters, market dynamics information, and user behavior characteristics. This represents the predicted heat load time series value for a predetermined future time period. These represent the heat source data and the pipeline topology diagram, respectively. Represents the spatial topology feature mapping value of the pipeline network. This represents the preset model error. These are the weighting coefficients.

7. The multi-heat source networked heating optimized operation system as described in claim 6, characterized in that, The control module includes: An economic cost item construction unit is used to construct a fuel cost function based on unit fuel cost and fuel consumption, and to obtain economic cost items based on the fuel cost function. An environmental cost item construction unit is used to construct a carbon emission cost function based on the carbon emission per unit of fuel and the fuel consumption, and to obtain environmental cost items based on the carbon emission cost function. A supply-demand deviation term construction unit is used to construct a supply-demand deviation term based on the final heat load forecast and the target heat supply. The target optimization unit is used to construct a target optimization function based on the economic cost item, the environmental cost item, and the supply and demand deviation item, and adjust the output value of each heat source based on the target optimization function to achieve a balance between the economy, environmental protection, and supply and demand matching of the heating system.

Images