Intelligent dispatching optimization method and system for energy-saving heating

Abstract

The present application relates to the technical field of wisdom energy management, and discloses an intelligent scheduling optimization method and system for energy-saving heating, which comprises the following steps: defining a regulation cost function for each heating partition; obtaining the marginal energy price of the system at each scheduling time; comparing the marginal energy price of the system with the current regulation cost function of each heating partition to determine the final decision setting value of the partition; and solving the collaborative control instruction set according to the final decision setting values of all heating partitions. The system comprises a central processing platform, a plurality of partition controllers and a communication network. The present application can convert the user thermal comfort into flexible resources that can be guided by the global price signal, and realize the global economic operation and energy consumption minimization of the heating system under the premise of ensuring the comfort by combining the distributed economic decision with the centralized physical collaboration.

Description

Technical Field

[0001] This invention relates to the field of smart energy management technology, and in particular to a smart scheduling optimization method and system for energy-saving heating. Background Technology

[0002] Urban centralized heating systems are crucial infrastructure for ensuring people's livelihoods and industrial production. Their operation is energy-intensive, yet they possess significant energy-saving potential. Traditional heating dispatching methods rely primarily on manual experience or simple automated control, resulting in a crude operating mode of "high flow rate, small temperature difference," leading to severe energy waste. In recent years, with the development of smart heating technology, advanced algorithms such as model predictive control and reinforcement learning have been introduced to optimize heat source output and pipeline regulation, aiming to reduce system energy consumption.

[0003] However, existing technical solutions generally have a fundamental limitation: they mainly focus on one-sided optimization on the supply side, that is, treating the user's heat load demand as a fixed input parameter or rigid constraint, and the scheduling goal is to reduce the production and distribution costs of heat sources as much as possible while meeting these given demands.

[0004] This supply-side centralized optimization paradigm fails to fully utilize the flexibility of the demand side. In reality, users' perception of thermal comfort is elastic within a certain range, and the thermal inertia varies significantly among different users and buildings. Existing scheduling systems lack an effective mechanism to quantify, aggregate, and transform the massive, dispersed user-side thermal comfort elasticity into system-schedulable and tradable virtual energy storage resources. Simultaneously, there is a lack of a unified price signal that can reflect the global scarcity of resources in real time, guiding these distributed resources to make spontaneous and economically optimal adjustments. Summary of the Invention

[0005] The purpose of this invention is to solve the problems of supply and demand separation and rigid scheduling in the existing technology, and to propose an intelligent scheduling optimization method and system for energy-saving heating.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A smart scheduling optimization method for energy-saving heating, used for the energy-saving and economical operation of district centralized heating systems, includes the following steps:

[0008] Step S1: Define an adjustment cost function for each heating zone in the heating network. The heating zone is a relatively independent control unit divided according to building type, insulation level or geographical location. The adjustment cost function maps the deviation of the user room temperature setpoint of the zone from its historical statistical benchmark value to the marginal cost change required for the heating system to achieve the deviation. The historical statistical benchmark value is the average room temperature calculated by weighting the normal operation data of the zone in at least one past heating season, after removing outliers.

[0009] Step S2: At each scheduling moment, obtain the system marginal energy price, which represents the global energy consumption cost of the system.

[0010] Step S3: For each heating zone, compare the marginal energy price of the system with the current adjustment cost function of that zone, and independently decide the final decision setpoint for that zone based on the comparison result.

[0011] Step S4: Based on the final decision set values ​​of all heating zones, solve for the coordinated control command set of heat source output, pipeline valve group and circulating pump that satisfies the hydraulic and thermal balance constraints of the heating network.

[0012] Furthermore, an intelligent scheduling and optimization system for energy-saving heating is provided to achieve intelligent scheduling and optimization for energy-saving heating. The system includes:

[0013] The central processing platform is used to calculate and publish the marginal energy price of the system, receive the room temperature setpoint decision results of each heating zone, and generate a set of coordinated control instructions based on the decision results of all zones.

[0014] Multiple zone controllers are deployed in each heating zone to store and call the adjustment cost function of their respective zone, receive the marginal energy price of the system, and execute comparison logic to output the final decision setpoint of the zone.

[0015] The communication network connects the central processing platform with all zone controllers and is used to transmit the system's marginal energy price, the final decision setpoints for each zone, and the collaborative control instruction set.

[0016] The beneficial effects of the technical solution provided by this invention include at least the following:

[0017] This invention quantifies and encapsulates user-side thermal comfort elasticity by introducing an adjustment cost function. This transforms unstructured user demands, which are difficult to handle in traditional scheduling, into flexible resources with clear economic significance that can be directly invoked by the system model. This lays the technical foundation for proactive demand-side management in scheduling.

[0018] This invention uses the marginal energy price of the system calculated iteratively based on the interaction of supply and demand as a globally unified signal. This provides a clear and consistent economic incentive for all distributed decision-making units, enabling countless local self-interested decisions to spontaneously approach the global economic optimum under the guidance of the price signal, thus solving the coordination problem of large-scale distributed systems.

[0019] This invention, through a distributed autonomous mechanism that compares prices with adjustment cost functions and makes decisions, enables each heating zone to make optimal room temperature setpoint adjustments in real time and independently based on its own characteristics and global signals. This achieves automated, real-time, and refined scheduling decisions, significantly reduces dependence on the computing power of the central optimizer, and the system has extremely high scalability.

[0020] This invention, by taking the results of all regional economic decisions as the tasks that the physical layer must satisfy and solving the collaborative control instructions, can ensure that the optimal strategy at the economic level is accurately and efficiently transformed into the actual actions of physical equipment, realizing the seamless connection from economic intelligence to physical control, and ultimately significantly reducing the overall energy consumption and cost of the system while ensuring the quality of heating.

[0021] Through the above-described overall technical solution, this invention can upgrade the heating system from a passive energy transmission and distribution network that responds to demand into a smart energy operation platform with self-adaptive and self-optimizing capabilities, providing core technical support for heating companies to achieve refined, intelligent, and low-carbon business operations. Attached Figure Description

[0022] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of the system configuration provided in an embodiment of the present invention. Detailed Implementation

[0025] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the intelligent scheduling optimization method and system for energy-saving heating proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0026] 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.

[0027] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0028] The specific scheme of the intelligent scheduling optimization method and system for energy-saving heating provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0029] Please see Figure 1 This document illustrates a flowchart of an intelligent scheduling optimization method for energy-saving heating provided by an embodiment of the present invention. This method is used for energy-saving and economical operation of a regional centralized heating system and includes the following steps:

[0030] Step S1: Define an adjustment cost function for each heating zone in the heating network. A heating zone is a relatively independent control unit divided according to building type, insulation level or geographical location. The adjustment cost function maps the deviation of the user room temperature setpoint of the zone from its historical statistical benchmark value to the marginal cost change required for the heating system to achieve the deviation. The historical statistical benchmark value is the average room temperature calculated by weighting the normal operation data of the zone in at least one past heating season, after removing outliers.

[0031] Step S2: At each scheduling moment, obtain the system marginal energy price, which represents the global energy consumption cost of the system.

[0032] Step S3: For each heating zone, compare the marginal energy price of the system with the current adjustment cost function of that zone, and independently decide the final decision setpoint for that zone based on the comparison result.

[0033] Step S4: Based on the final decision set values ​​of all heating zones, solve for the coordinated control command set of heat source output, pipeline valve group and circulating pump that satisfies the hydraulic and thermal balance constraints of the heating network.

[0034] It should be noted that heating zones refer to relatively independent control units divided according to characteristics such as building type, insulation level, and geographical location. Each unit has its own unique operating characteristics and its range is determined by physical boundaries or logical division methods. This is mainly to adapt to the heating differences in different areas and achieve precise scheduling.

[0035] The adjustment cost function is a mathematical expression that quantifies the relationship between the deviation of room temperature and the change in marginal cost. It is constructed by fitting and constraining historical data. The function parameters of different zones are calibrated independently. The main purpose is to clarify the economic cost of room temperature regulation and provide a quantitative basis for zoning decisions.

[0036] Historical statistical benchmark values ​​refer to the average room temperature values ​​obtained through statistical analysis based on normal operating data of multiple heating seasons in the past. These values ​​are calculated by weighting the average after removing outlier data. The main purpose is to establish a generally accepted comfortable room temperature benchmark for users, serving as a reference for calculating adjustment ranges.

[0037] Marginal cost change refers to the additional comprehensive cost incurred by the system to make the room temperature deviate from the benchmark value by one unit. It includes fuel consumption, equipment wear and tear, operation and maintenance expenses, etc. It is calculated through cost decomposition and conversion formula, mainly to establish a direct link between room temperature regulation and economic cost.

[0038] The scheduling time refers to a preset fixed time interval node, which is set to 15 minutes or 30 minutes according to the system response requirements. The scheduling process is triggered synchronously by the clock, mainly to ensure the regularity and timeliness of the scheduling and adapt to the dynamic changes in heat load.

[0039] The marginal energy price of a system refers to the minimum economic cost required for the system to meet an additional unit of heat load at the current moment. It is calculated by taking into account the energy consumption cost of the entire process. Its main purpose is to provide a unified economic decision-making benchmark for the entire network and to coordinate regional regulation behavior with overall objectives.

[0040] The final decision setpoint refers to the target room temperature value determined within the allowable range based on a comparison of price and cost for each zone. It is obtained by adapting the theoretical optimal solution to the constraints, mainly to balance economic costs and user comfort needs.

[0041] Hydraulic and thermal balance constraints refer to the core conditions for ensuring the safe and stable operation of the pipeline network. Hydraulic balance requires reasonable flow distribution, and thermal balance requires temperature transmission to meet standards. These are defined by physical equations and operating standards, mainly to avoid network imbalance caused by local regulation.

[0042] The coordinated control instruction set refers to a collection of operating parameters such as heat source output, valve opening, and pump frequency. These parameters are matched with each other and meet the constraints. The main purpose is to provide directly executable operational guidelines to ensure the implementation of the scheduling plan.

[0043] In one specific implementation, the district centralized heating system covers a heating area of ​​600,000 square meters and is divided into 8 heating zones according to building type, including 5 residential zones and 3 commercial building zones.

[0044] In step S1, operational data for the past three heating seasons for each zone is collected, including indoor and outdoor temperatures, supply and return water parameters, energy consumption records, and room temperature setpoints. Data collection is performed every 10 minutes, synchronously acquired via industrial sensors and a data acquisition platform. After cleaning and removing outliers, a weighted average method is used to calculate the historical statistical baseline value for each zone. The baseline value for residential zones is between 19℃ and 20℃, and for commercial building zones it is between 20℃ and 21℃. A gradient boosting tree algorithm is selected, using the room temperature deviation (-5℃ to 5℃) as input and the corresponding incremental comprehensive cost as output, to fit the initial mapping relationship. A monotonicity regularization penalty term is added during training to ensure that the cost increases monotonically with the deviation. Finally, the cost is smoothed using a moving average method to obtain the adjustment cost function for each zone. For example, the function for a residential zone is marginal cost = 0.8 × squared deviation + 0.3 × deviation + 0.5.

[0045] In step S2, the scheduling interval is set to 15 minutes. At each scheduling time, the outdoor temperature for the next moment is obtained from the weather forecast. Combined with the current time period (peak, average, and valley), historical prices for similar operating conditions are retrieved from the historical price model database as initial forecast values. Based on these initial forecast values, adjustment decisions for each zone are simulated, and the total predicted heat load is obtained. Combining the operating cost curves and start-up / shutdown amortization costs of the three gas-fired heat sources within the system, the unit cost of the marginal heat source required to meet the load is calculated as a candidate price. The initial forecast value and the candidate price are compared. If the difference is less than 3%, it is determined as the final system marginal energy price; otherwise, the candidate price is used as the new initial value for iterative calculation, with an upper limit of 5 iterations.

[0046] When executing step S3, for each zone, if the adjustment cost function is in analytical form, the equation is directly solved to obtain the theoretically optimal room temperature. If it is a black-box model, a bisection method is used to search within the range of 16℃ to 25℃, with an accuracy controlled within 0.1℃. Combining the user's contracted minimum temperature (18℃), the current outdoor temperature, and the building's insulation level, the allowable room temperature range is determined: Allowable lower limit = max(contracted minimum temperature, outdoor temperature + safety temperature difference), with the safety temperature difference set at 5℃ to 8℃ according to the insulation level, and the allowable upper limit set at 24℃. The theoretically optimal value is mapped to the allowable range to obtain the final decision setting value. For example, the theoretically optimal value for a commercial zone is 24.5℃, and the final decision setting value is 24℃.

[0047] In step S4, the overall heat load distribution is calculated based on the final decision setpoints for each zone, the heating area, and the building thermal index. An optimization model is established, incorporating network flow conservation, temperature transmission laws, and equipment operating limitations. This model is decomposed into sub-problems and solved in parallel using the alternating direction multiplier method. A coupling sensitivity matrix is ​​constructed to quantify the relationship between zone load and network flow. During the iteration process, the flow distribution scheme is corrected. Iteration stops when the parameter deviation is less than 0.5%, and a coordinated control instruction set is output, including the heating capacity of each heat source (ranging from 60 GJ / h to 90 GJ / h), the opening degree of network valve groups (ranging from 50% to 80%), and the operating frequency of circulating pumps (ranging from 40 Hz to 50 Hz), which is then sent to the corresponding equipment for execution.

[0048] Step S1 further includes the following sub-steps:

[0049] S1-1: For each heating zone in the heating network, collect historical operating data for that zone. The operating data includes indoor and outdoor temperatures, valve opening, supply and return water temperatures, changes in user room temperature setpoints, and corresponding changes in total system energy consumption.

[0050] S1-2 takes the change in the user's room temperature setpoint as input and the marginal cost calculated from the corresponding change in the total system energy consumption as output, and fits the mapping relationship between the input and output through a machine learning algorithm;

[0051] S1-3, apply monotonic constraints to the fitted mapping relationship to ensure that the mapping relationship is a monotonically increasing function;

[0052] S1-4, normalize and smooth the processed monotonically increasing function to obtain the adjustment cost function of the heating zone.

[0053] Furthermore, in sub-steps S1-3, the steps to ensure that the mapping relationship is a monotonically increasing function include:

[0054] During the model training phase of a machine learning algorithm, a monotonicity regularization penalty term is added to the loss function. The monotonicity regularization penalty term L_mono is calculated according to the following formula: L_mono=λ*Σ_imax(0,-∇f(x_i)), where λ is the regularization strength coefficient, Σ_i represents the summation over training sample index i, f(x_i) is the model's predicted output at input x_i; ∇f(x_i) is the gradient of the model's predicted output f at x_i with respect to input x, and the max(0,-∇f(x_i)) function ensures that only negative gradients are penalized.

[0055] By optimizing the loss function with a monotonicity regularization penalty, the trained mapping relationship maintains that the output value monotonically increases with the input value within its domain.

[0056] It should be noted that collecting historical operating data for this zone refers to collecting key operating parameters and energy consumption data related to the normal operation of the heating zone in the past. This data is continuously acquired through sensors and data acquisition terminals deployed in the pipeline network, equipment, and user terminals. The main purpose is to accumulate the basic data required to build the mapping relationship and ensure that the data covers different operating conditions and scenarios.

[0057] Indoor and outdoor temperatures refer to the actual indoor temperature of users within a zone and the outdoor ambient temperature. These are collected and recorded periodically by temperature sensors, primarily to reflect the environmental and user-side temperature background and provide environmental variable references for energy consumption change analysis.

[0058] Valve opening degree refers to the degree to which a regulating valve in a pipeline network is open. It is collected by the valve's built-in position sensor and is mainly used to reflect the pipeline network's flow regulation status, and to correlate room temperature changes with system regulation behavior.

[0059] Supply and return water temperatures refer to the temperatures of the heat transfer medium in the supply and return water pipes of the heating network. These temperatures are collected by temperature sensors embedded in the pipes and are mainly used to reflect the heat transfer status of the network, serving as a key parameter for energy consumption calculation.

[0060] The change in the user's room temperature setpoint refers to the difference between the user's adjusted target room temperature value and the original setpoint. This value is obtained through the operation records of the temperature control device and is mainly used to quantify the user's adjustment needs as an input variable for the mapping relationship.

[0061] The corresponding change in total system energy consumption refers to the total amount of additional energy consumed by the system due to the adjustment of the user's room temperature setting. It is collected and calculated by energy metering equipment, mainly to clarify the relationship between room temperature regulation and energy consumption, and to provide a basis for marginal cost calculation.

[0062] Using the change in the user's room temperature setpoint as input and the marginal cost as output means establishing a quantitative correspondence between the two. The input variable reflects the degree of adjustment, and the output variable reflects the economic cost. The main purpose is to construct a correlation model between cost and adjustment behavior through algorithm fitting.

[0063] Fitting mapping relationships using machine learning algorithms refers to selecting algorithms adapted to nonlinear data, training and learning on input and output data, and exploring the potential correlation between the two. The main purpose is to obtain a mathematical mapping relationship that can accurately quantify adjustment costs.

[0064] Monotonicity constraint processing of the mapping relationship refers to using technical means to force the mapping relationship to meet the law that the output increases with the input, so as to ensure that the larger the room temperature adjustment range, the higher the marginal cost. This is mainly to conform to the physical logic and economic laws of the operation of the heating system.

[0065] Normalization and smoothing refer to the standardization and fluctuation elimination of constrained functions, so that the function values ​​are of a uniform magnitude and the curves are continuous and stable. This is mainly to improve the practicality and stability of the function and facilitate subsequent decision-making calculations.

[0066] Obtaining the regulation cost function for each heating zone refers to the final formation of a quantitative model that can be directly used for calculation. This function is specific to each zone and reflects its unique operating characteristics, mainly to provide accurate cost basis for zone room temperature regulation decisions.

[0067] Adding a monotonicity regularization penalty term to the loss function refers to adding an extra constraint term to the error calculation stage of model training. This penalty applies to prediction results that do not conform to the monotonically increasing law, mainly to guide the direction of model training and ensure that the output results conform to the preset law.

[0068] Quantifying the negative gradient value of the model's predicted output with respect to the input refers to identifying situations where the model output decreases as the input increases, and calculating the penalty strength according to set rules. This is mainly to specifically correct model biases and avoid unreasonable situations where costs and adjustment magnitudes change inversely.

[0069] Optimizing a loss function with a monotonicity regularization penalty term involves iteratively adjusting model parameters through algorithms to minimize prediction errors while satisfying monotonicity constraints. This is primarily aimed at balancing the model's fitting accuracy and conformity to patterns, thereby obtaining a reliable mapping relationship.

[0070] In one specific implementation, step S1 is carried out for six heating zones in a certain city.

[0071] When executing S1-1, temperature sensors, flow sensors, valve position sensors, and energy metering devices are deployed in each zone to collect historical operating data from the past three heating seasons. The data collection interval is set to 10 minutes. The collected data includes indoor temperature, outdoor temperature, pipeline valve opening, supply water temperature, return water temperature, user room temperature setpoint adjustment records, and corresponding system natural gas consumption and electricity consumption for each time period of the day. The collected data is preprocessed to remove outliers caused by sensor malfunctions and invalid data during downtime, retaining approximately 120,000 valid data entries per zone.

[0072] When executing S1-2, the change in the user's room temperature setpoint is used as the input variable, and the change in the total system energy consumption, converted into marginal cost based on the energy unit price, is used as the output variable. The input variable range for each partition is -5 to 5 degrees Celsius, and the output variable ranges from 0.2 to 3.5 yuan per degree Celsius. A random forest algorithm is selected as the machine learning algorithm. The effective data for each partition is divided into training and test sets in a 7:3 ratio. The training set data is used to train the model, fitting the mapping relationship between the input and output. The test set is used to verify the model's fitting accuracy, ensuring that the average error is below 5%.

[0073] In S1-3, during the model training phase of the random forest algorithm, a monotonicity regularization penalty term is added to the loss function, with a weight of 0.15. This penalty term calculates the negative gradient of the model's predicted output with respect to the input. For each instance where the output decreases as the input increases, a corresponding proportional penalty is added to the loss value. The loss function with this penalty term is optimized using the gradient descent algorithm. After 500 iterations of training, a mapping relationship that satisfies the monotonicity requirement is obtained. Verification shows that the mapping relationship for all partitions satisfies the condition that the output monotonically increases with the input.

[0074] When executing S1-4, the max-min normalization method is used to normalize the processed monotonically increasing function, mapping the function value to the interval between 0 and 1, thus eliminating the influence of differences in data magnitude between different zones. Subsequently, the moving average method is used to smooth the normalized function, setting the moving window size to 5 to eliminate function curve jitter caused by data fluctuations. Finally, the regulation cost function for each heating zone is obtained, which can be directly used for subsequent cost calculation and decision-making for zone room temperature regulation.

[0075] Step S2 further includes the following sub-steps:

[0076] S2-1: Based on the predicted outdoor temperature and time period of the next scheduling time, retrieve the historical marginal price under similar operating conditions from the pre-trained historical price pattern library and use it as the initial system marginal energy price prediction value.

[0077] S2-2. Based on the current system marginal energy price forecast and the adjustment cost function of each heating zone, the room temperature setpoint adjustment decision of each heating zone at the next scheduling time is predicted in parallel, and the total heat load of the predicted system is obtained by summarizing.

[0078] S2-3. Based on the predicted total heat load of the system and the current operating conditions, and combined with the average temperature rise-cost mapping relationship of the entire network fitted by the historical data of the system, calculate the marginal cost corresponding to the load change, which serves as the candidate marginal energy price for this round.

[0079] S2-4: Calculate the difference between the candidate marginal energy price and the current system marginal energy price forecast. If the difference is less than the preset convergence threshold, publish the candidate marginal energy price as the final system marginal energy price.

[0080] S2-5, if the difference is greater than or equal to the preset convergence threshold, then the candidate marginal energy price is used as the new round of system marginal energy price prediction value, and the process returns to step S2-2 for iteration.

[0081] It should be noted that the predicted outdoor temperature for the next scheduling time refers to the prediction of the outdoor ambient temperature during the upcoming scheduling cycle. It is calculated by combining meteorological forecast data with historical temperature patterns for the same period. This is mainly to adapt to the impact of ambient temperature on heat load in advance and to provide environmental parameter support for price forecasting.

[0082] The time period refers to the time interval in which the next dispatch time is located. It is divided according to the peak-valley-flat characteristics of electricity consumption or the heat consumption patterns of users. This is mainly to distinguish the energy consumption characteristics of different time periods and make the historical operating conditions retrieved more targeted.

[0083] A pre-trained historical price pattern library refers to a database that stores different operating conditions and their corresponding historical marginal prices. It is constructed by classifying historical data by operating conditions through clustering algorithms. The main purpose is to quickly retrieve price references for similar scenarios and provide a data source for initial prediction values.

[0084] Historical marginal prices under similar operating conditions refer to the marginal prices of past scheduling times that have a high degree of matching with the current predicted outdoor temperature, time period, and other conditions. They are obtained by filtering through operating condition similarity algorithms, mainly to ensure that the initial predicted values ​​have reasonableness and reference value.

[0085] The initial marginal energy price forecast of the system refers to the price estimate that serves as the starting point for iterative calculations. It is determined based on historical similar operating conditions and is mainly intended to provide an initial benchmark for iterative calculations, thereby driving the price to gradually converge to the optimal value.

[0086] Parallel simulation prediction of room temperature setpoint adjustment decisions for each heating zone refers to simulating and calculating the adjustment behavior of all zones simultaneously, without processing them sequentially. This is mainly to improve prediction efficiency and adapt to the scheduling timeliness requirements of multi-zone systems.

[0087] The total predicted system heat load is obtained by summing the predicted heat load demands of all zones to obtain the total heat load data at the network level. This is mainly to reflect the overall heat demand of the system and provide a basis for calculating heat source costs.

[0088] The start-up and shutdown costs of each heat source refer to the one-time costs incurred when starting or stopping the heat source equipment, including preheating consumption, equipment wear and tear, etc. They are calculated based on actual operation and maintenance data, mainly to comprehensively calculate the overall cost of heat source operation and avoid ignoring the cost impact of the start-up and shutdown process.

[0089] The operating cost curve is a curve that describes the relationship between the heat supply of a heat source and the unit operating cost. It is obtained by fitting heat source characteristic tests and historical data. Its main purpose is to quantify the continuous operating cost under different heat supply conditions and support the calculation of marginal cost.

[0090] The last marginal heat source refers to the heat source that is put into operation or increases its output when the predicted total heat load is met. Its cost is usually higher than that of the heat sources put into operation earlier. It is mainly used to determine the marginal cost of the system meeting the additional load, and serves as the core basis for candidate marginal energy prices.

[0091] The unit comprehensive cost refers to the unit heat load cost calculated by summing the start-up and shutdown costs of the marginal heat source with the operating costs. It comprehensively reflects the cost of using the heat source throughout its entire life cycle, mainly to accurately reflect the economic cost of marginal heating and ensure the comprehensiveness of candidate prices.

[0092] Calculating the difference between candidate marginal energy prices and current forecasts involves obtaining the difference and proportion of the two through numerical calculations, quantifying the magnitude of price iteration changes, and mainly to determine whether prices have stabilized, providing quantitative indicators for convergence judgment.

[0093] The preset convergence threshold is the critical value at which price iterations reach a stable state. It is set according to the system scheduling accuracy requirements, mainly to avoid meaningless iterative loops and ensure the efficiency and accuracy of price calculation.

[0094] The release of the final system marginal energy price means that the converged candidate price is determined as the basis for formal decision-making and distributed to each region for adjustment reference. The main purpose is to provide a unified economic benchmark for the entire network and guide precise decision-making in each region.

[0095] Returning to step S2-2 for iteration means using the unconverged candidate price as a new initial prediction value and repeating the process of simulation decision-making, load and price calculation. This is mainly to gradually correct the price deviation so that the final price is closer to the actual load demand and system cost characteristics.

[0096] In one specific implementation, when executing step S2, the scheduling cycle is first determined to be 15 minutes, and the next scheduling time is the morning rush hour on a winter workday.

[0097] When executing S2-1, the meteorological department's forecast outdoor temperature for the next moment is obtained as -3 degrees Celsius. Based on the corresponding morning peak period, similar operating conditions are retrieved from the historical price model database. Historical data with outdoor temperatures between -4 and -2 degrees Celsius and belonging to the same morning peak period are selected, and the corresponding historical marginal price of 2.7 yuan per degree Celsius is used as the initial system marginal energy price forecast value.

[0098] When executing S2-2, based on the initial predicted value of 2.7 yuan per degree Celsius, the adjustment cost function that has been built for each heating zone is called. The room temperature setpoint adjustment decision of each zone is simulated simultaneously through the parallel computing framework. That is, each zone solves the optimal room temperature when the adjustment cost is equal to the predicted price. Then, combined with its own constraints, the final adjustment direction and magnitude are determined. The heat load demand changes of all zones are summarized to obtain the predicted total heat load of the system as 110 degrees Celsius per hour.

[0099] When executing S2-3, the system contains 3 gas heat sources. Based on the predicted total heat load demand of the entire network (110 degrees Celsius per hour), the output scheme of each heat source is determined as follows: heat source 1 at full load (60 degrees Celsius per hour), heat source 2 at full load (40 degrees Celsius per hour), and heat source 3 as a marginal heat source (10 degrees Celsius per hour).

[0100] Calculating the marginal energy price of the candidate system: The pre-defined "average temperature rise of the entire network - cost mapping relationship" indicates that under the current operating conditions (e.g., current outdoor temperature -3℃, average network efficiency 0.92, system in the morning peak period), when heat source 3 is used as a marginal heat source and its output is increased by 1 degree Celsius per hour, the resulting change in the system's marginal cost is 4.288 yuan / (degree Celsius per hour). Therefore, to meet the additional marginal heat load of 10 degrees Celsius per hour at this scheduling moment, the corresponding marginal energy price of the candidate system (i.e., the "marginal cost of increasing the average room temperature of the entire network by 1℃") is calculated as follows: Marginal energy price of the candidate system = unit output cost of the marginal heat source × marginal output = 4.288 yuan / (degree Celsius per hour) × 10 degrees Celsius per hour = 42.88 yuan / degree Celsius; this 42.88 yuan / degree Celsius is the marginal energy price of the candidate system calculated in this round.

[0101] When executing S2-4, the preset convergence threshold is 3%. The difference ratio between the candidate marginal energy price of 42.88 yuan per degree Celsius and the initial predicted value of 2.7 yuan per degree Celsius is calculated. The difference ratio is far greater than 3%, and the convergence requirement is not met.

[0102] In step S2-5, the candidate marginal energy price of 42.88 yuan per degree Celsius is used as the new predicted system marginal energy price. The process returns to step S2-2 to resimulate the decisions for each zone, summarizing the results to obtain a new predicted total system heat load of 95 degrees Celsius per hour. The candidate marginal energy price is then recalculated as 38.5 yuan per degree Celsius. After repeating this iteration three times, the difference between the candidate marginal energy price and the current predicted value decreases to 2.5%, which is less than the convergence threshold. Therefore, the candidate marginal energy price is determined as the final system marginal energy price and published.

[0103] Step S3 further includes the following sub-steps:

[0104] S3-1, For the heating zone currently being processed, a theoretically optimal room temperature setpoint is obtained based on its adjustment cost function, such that the marginal cost represented by the adjustment cost function is equal to the marginal energy price of the system at this setpoint.

[0105] S3-2, Based on the user's thermal comfort contract and the current indoor and outdoor conditions, obtain the allowable room temperature setpoint range for this heating zone;

[0106] S3-3, according to the predefined constraint projection rules, the theoretical optimal room temperature setpoint is mapped to the range of allowable room temperature setpoints to obtain the final decision setpoint.

[0107] Furthermore, in sub-step S3-1, a theoretically optimal room temperature setpoint is obtained in the following way:

[0108] Obtain the historical statistical baseline room temperature T_base for the current heating zone;

[0109] When the adjustment cost function is in analytical form, the optimal deviation range x_opt is obtained by solving the equation f(x)=P, and the theoretical optimal room temperature setpoint T_opt is calculated by the formula T_opt=T_base+x_opt, where f(x) is the adjustment cost function and P is the marginal energy price of the system.

[0110] When the adjustment cost function is a discrete lookup table or a black box model, a value of x is found in the preset interval using the bisection method or the golden section method, such that the absolute value of the difference between f(x) and P is less than the preset accuracy threshold, and this value of x is used as the theoretically optimal room temperature setting value.

[0111] Furthermore, in sub-step S3-2, the range of room temperature setpoint is allowed to be defined by a lower limit value T_min and an upper limit value T_max;

[0112] The lower limit value T_min is dynamically determined according to the following rule: T_min = max(T_contract, T_outdoor + ΔT_safe), where T_contract is the absolute minimum temperature specified in the user's thermal comfort contract, T_outdoor is the current outdoor temperature, and ΔT_safe is the safety temperature difference for preventing pipe freezing determined according to the building insulation performance. ΔT_safe is determined according to the building insulation performance level. The higher the insulation performance level, the smaller the value of ΔT_safe, and the value range is 5°C - 8°C;

[0113] The upper limit value T_max is a preset fixed constant or a fixed value specified by the user's thermal comfort contract.

[0114] Furthermore, in sub-step S3-3, the predefined constraint projection rule is:

[0115] Compare the theoretical optimal room temperature setting value x_opt with the lower limit value T_min and the upper limit value T_max of the allowable room temperature setting range;

[0116] If x_opt < T_min, set the final decision setting value to the lower limit value;

[0117] If x_opt > T_max, set the final decision setting value to T_max;

[0118] If T_min ≤ x_opt ≤ T_max, set the final decision setting value to the theoretical optimal room temperature setting value.

[0119] It should be noted that the theoretical optimal room temperature setting value refers to the room temperature target value that makes the marginal cost equal to the system marginal energy price only from the perspective of economic cost, without considering other constraints. It is mainly to obtain an ideal economic optimal solution as the basis reference for subsequent decisions.

[0120] The marginal cost represented by the regulation cost function refers to the unit regulation cost corresponding to a specific room temperature setting value output by this function, which reflects the economic cost brought by the change of room temperature. It is mainly to establish an equivalent comparison relationship with the system marginal energy price to achieve the matching of cost and price.

[0121] The system marginal energy price refers to the benchmark price determined finally in step S2, which represents the global energy consumption cost of the system. It is mainly to provide a unified economic decision-making scale to ensure the consistency of the partition decision and the global goal.

[0122] The regulation cost function in analytical form refers to a function that can be expressed by a clear mathematical expression and has the condition of direct solution. It is mainly to quickly obtain the theoretical optimal solution through algebraic operations and improve the decision-making efficiency.

[0123] The adjustment cost function in discrete lookup table form refers to storing the marginal cost corresponding to different room temperatures in the form of key-value pairs. There is no unified mathematical expression. It is mainly used to adapt to scenarios that cannot be fitted as continuous functions. The solution is achieved by looking up the table and interpolation.

[0124] The adjustment cost function in the form of a black box model refers to a model whose results can only be obtained through the input-output relationship, and whose internal logic is not visible. It is mainly used to deal with cost mapping under complex working conditions and to approximate the optimal solution through numerical algorithms.

[0125] The bisection method or golden section method is a numerical solution algorithm based on the principle of gradually narrowing the interval. It has the characteristics of fast convergence speed and small amount of computation. It is mainly used to efficiently find the optimal value that meets the accuracy requirements when there is no analytical expression.

[0126] The preset range refers to the predefined search range of room temperature, which is determined by combining user comfort needs and equipment operating limits. It is mainly to limit the solution boundary and avoid the search range being too large, which would lead to low computational efficiency or unreasonable results.

[0127] The preset accuracy threshold refers to the critical error value for determining whether the solution result meets the standard. It is set according to the accuracy requirements of scheduling and control, mainly to balance the calculation accuracy and efficiency, and to ensure that the result meets the actual application requirements.

[0128] A user thermal comfort contract is an agreement signed between the heating provider and the user that clearly defines the room temperature guarantee standard. It stipulates core terms such as the minimum comfort temperature and is mainly intended to clarify the bottom line of the user's rights and ensure that the permissible room temperature range complies with the contract.

[0129] The current indoor and outdoor status refers to the actual indoor temperature and outdoor ambient temperature of the user in the zone at the current moment. It is collected in real time by sensors, mainly to dynamically adapt to environmental changes and ensure the rationality and safety of the allowable room temperature range.

[0130] The permissible room temperature setpoint range refers to the feasible room temperature range that takes into account user comfort, equipment safety, and contractual agreements. It is limited by both upper and lower limits, mainly to provide a constraint boundary for the theoretical optimal value and avoid decisions that exceed the feasible range.

[0131] The lower limit value T_min refers to the lowest critical value of the room temperature that can be allowed. It is dynamically determined in combination with contract requirements and antifreeze safety. It is mainly to ensure the basic comfort of users and prevent the pipes from freezing, so as to avoid safety hazards caused by low temperature.

[0132] The absolute minimum temperature T_contract refers to the lowest room temperature that the user can accept as clearly stipulated in the contract. It is the core guarantee of the user's rights and interests, mainly to ensure that the decision does not violate the contract and to avoid performance risks.

[0133] The safe temperature difference ΔT_safe to prevent pipe freezing refers to the difference between the outdoor temperature and the lowest indoor temperature, determined based on the building's thermal insulation performance. The worse the insulation performance, the larger the difference. It is mainly to compensate for the indoor temperature and prevent the pipe from being damaged by freezing due to the low outdoor temperature.

[0134] The upper limit value T_max refers to the highest critical value of room temperature that can be allowed. It is set based on comfort experience and energy-saving goals, mainly to avoid energy waste and user discomfort caused by excessively high room temperature.

[0135] Predefined constraint projection rules refer to explicit rules that map theoretical optimal values ​​to feasible intervals. They are implemented through simple numerical comparisons and are mainly used to quickly handle situations where theoretical optimal values ​​exceed constraints, thus obtaining a final decision that balances economy and feasibility.

[0136] In one specific implementation, when performing step S3, a residential heating zone in a certain city is selected as the processing object. The adjustment cost function of this zone is in analytical form f(x) = 0.8x² + 0.3x + 0.5, and the marginal energy price of the system is P = 3.2 yuan per degree Celsius.

[0137] When executing S3-1, since the adjustment cost function is in analytical form, the equation 0.8x² + 0.3x + 0.5 = 3.2 is solved directly. Algebraic operations yield x = 1.8 degrees Celsius (negative solutions are discarded). Combined with the historical baseline value of 20 degrees Celsius for this region, the theoretically optimal room temperature setting is 21.8 degrees Celsius. If the adjustment cost function for this region is a black-box model, the preset room temperature search range is 16 to 26 degrees Celsius. A bisection method is used for iterative solving, with a preset accuracy threshold of 0.1 degrees Celsius. After 8 iterations, a x value of 21.7 degrees Celsius is found that satisfies the absolute value of the difference between f(x) and 3.2 being less than 0.1, and this is taken as the theoretically optimal room temperature setting.

[0138] When executing S3-2, the user's thermal comfort contract is consulted, and the absolute minimum temperature T_contract = 18 degrees Celsius is determined. The current outdoor temperature T_outdoor = -4 degrees Celsius is obtained through the sensor. The building insulation level of this zone is Class A, and the corresponding safe temperature difference to prevent pipe freezing is ΔT_safe = 5 degrees Celsius. T_outdoor + ΔT_safe = 1 degree Celsius is calculated. According to the rule T_min = max(18, 1) = 18 degrees Celsius; the preset upper limit value T_max = 24 degrees Celsius. Therefore, the allowable room temperature setting range for this zone is 18 to 24 degrees Celsius.

[0139] When executing S3-3, according to the predefined constraint projection rules, the theoretical optimal room temperature setpoint of 21.8 degrees Celsius is compared with the upper and lower limits of the allowable range. Since 21.8 degrees Celsius falls between 18 and 24 degrees Celsius, it is used as the final decision setpoint for this zone. If the theoretical optimal value is 17 degrees Celsius, which is lower than the lower limit of 18 degrees Celsius, then the final decision setpoint is 18 degrees Celsius; if the theoretical optimal value is 25 degrees Celsius, which is higher than the upper limit of 24 degrees Celsius, then the final decision setpoint is 24 degrees Celsius.

[0140] Step S4 further includes the following sub-steps:

[0141] S4-1 determines the heat load distribution of the entire network by taking the final decision setpoints of all heating zones as the heat load requirements that must be met.

[0142] S4-2, calculate the first derivative of the adjustment cost function of each heating zone at its final decision setpoint, which is used as the internal sensitivity of the zone's heat demand to its own flow rate change. Combine the internal sensitivities of all zones in the entire network into a diagonal matrix, and multiply it with the constant matrix characterizing the hydraulic resistance of the pipeline network to obtain the coupling sensitivity matrix.

[0143] S4-3 employs a distributed optimization algorithm to iteratively solve for the coordinated control commands that satisfy the hydraulic and thermal balance. In each iteration, the coupling sensitivity matrix is ​​used to perform feasibility prediction and direction correction on the calculated flow distribution scheme.

[0144] S4-4: When the iterative solution converges to the preset accuracy, output the cooperative control instruction set.

[0145] It should be noted that the required heat load refers to the total amount and distribution of heat that the system must guarantee, derived from the final decision setpoints of each zone. It is the core objective of the scheduling command solution, mainly to ensure that users' comfort needs are met and to avoid heat load deficits caused by optimization.

[0146] The overall heat load distribution refers to the spatial allocation data of the heat load of the entire system, which is formed by integrating the heat load demand of each zone according to its spatial location and pipeline topology. The main purpose is to clarify the load pressure of each branch of the pipeline and provide a basis for flow distribution.

[0147] The coupling sensitivity matrix is ​​a matrix that quantifies the degree of mutual influence between the heat load changes in each zone and the flow distribution in the pipeline network. The matrix elements characterize the correlation between the load changes in a single zone and the flow in other zones. The main purpose is to accurately capture the coupling relationship between zone load and pipeline network flow, and to provide quantitative support for iterative correction.

[0148] The change in heat load demand in each zone refers to the difference between the heat load corresponding to the final decision setpoint of the zone and the original load. It reflects the dynamic adjustment range of the load and is mainly used to clarify the degree of disturbance of the pipeline network caused by the load change, serving as an input variable for sensitivity calculation.

[0149] The degree of influence of pipeline flow distribution refers to the proportion of pipeline flow adjustment caused by changes in zoned load. It is calculated by the load-flow correlation derived from the adjustment cost function. The main purpose is to quantify the strength of the coupling effect and ensure that the matrix can truly reflect the system characteristics.

[0150] Distributed optimization algorithms are algorithms that decompose a global optimization problem into multiple local subproblems and achieve global optimality through parallel solving of subproblems and information exchange. They are mainly used to improve solution efficiency and adapt to the real-time scheduling needs of large-scale heating networks.

[0151] The coordinated control command that satisfies hydraulic and thermal balance refers to the operation command that simultaneously conforms to the flow conservation, pressure balance and temperature transmission law of the pipeline network. The parameters of each command are matched with each other, mainly to ensure the safe and stable operation of the pipeline network and avoid hydraulic imbalance or thermal failure.

[0152] Feasibility prediction refers to using the coupling sensitivity matrix to predict whether the current flow allocation scheme will lead to local overflow, underflow or temperature imbalance in the pipeline network, and to identify the defects of the scheme in advance. The main purpose is to reduce invalid iterations and improve the solution convergence speed.

[0153] Directional correction refers to adjusting the optimization direction of flow distribution based on feasibility prediction results, correcting parameters that exceed constraints, and ensuring that the scheme always iterates in the direction that satisfies hydraulic and thermal balance. This is mainly to avoid the iteration process from deviating from the constraint boundary and to ensure the feasibility of the final command.

[0154] The convergence of the iterative solution to the preset accuracy means that the difference between the control command parameters calculated in two consecutive iterations during the iteration process is less than the set threshold. This indicates that the algorithm has found a stable optimal solution. This is mainly to avoid excessive iteration that consumes computing resources, while ensuring the stability and accuracy of the commands.

[0155] Output collaborative control instruction set refers to organizing the converged optimization results into a set of executable instructions containing parameters such as heat source output, valve opening, and pump frequency, and sending them to the corresponding equipment. The main purpose is to transform the optimization scheme into actual operation and ensure that the scheduling decision is implemented.

[0156] In one specific implementation, when executing step S4, the system includes 10 heating zones and 3 main pipeline branches. When executing S4-1, the final decision setpoints of each zone are combined with its building heating area and insulation characteristics to calculate the heat load demand of each zone. The heat load of the residential zone is between 10 and 15 degrees Celsius per hour, and the heat load of the commercial zone is between 15 and 20 degrees Celsius per hour. After summing, the total heat load of the entire network is 130 degrees Celsius per hour. Then, according to the pipeline topology, the heat load of each zone is distributed to the corresponding branches to determine the heat load distribution of the entire network: 45 degrees Celsius per hour for main pipeline A, 50 degrees Celsius per hour for branch B, and 35 degrees Celsius per hour for branch C.

[0157] When executing S4-2, the steps for calculating the coupling sensitivity matrix are as follows:

[0158] 1. Calculate the internal sensitivity of each partition.

[0159] For each heating zone, calculate the first derivative of its regulation cost function at its final decision setpoint. This derivative value characterizes the instantaneous rate of change of its own heat load demand (in degrees Celsius per hour) when the room temperature setpoint of that zone changes slightly, i.e., its internal sensitivity. For example, if the regulation cost function of a zone is f(x) = 0.8x² + 0.3x + 0.5, and its final decision setpoint is 21.8℃, then the first derivative at that point is f'(21.8) = 1.6 * 21.8 + 0.3 = 35.18. This value of 35.18 is the internal sensitivity s_i of that zone.

[0160] 2. Construct the internal sensitivity diagonal matrix

[0161] Using the internal sensitivities s_1, s_2, ..., s_{10} of the 10 heating zones in the entire network as diagonal elements, construct an internal sensitivity diagonal matrix S: S=diag(s_1,s_2,...,s_{10}).

[0162] 3. Obtain the hydraulic resistance constant matrix of the pipeline network.

[0163] Based on the pre-calculated or identified pipeline topology and resistance parameters, a pipeline hydraulic resistance constant matrix H (10×10) is formed. This matrix is ​​symmetric, and its element h_ij represents the influence coefficient (unit: Pa / (m³ / h)) of the flow rate change in zone j on the pressure of node i in zone i.

[0164] 4. Calculate the coupling sensitivity matrix

[0165] By multiplying the internal sensitivity diagonal matrix S with the network hydraulic resistance constant matrix H, the final coupling sensitivity matrix A is obtained: A = S × H, and the dimension of matrix A is 10 × 10. The physical meaning of its element a_ij is: when the heat load demand (degrees Celsius per hour) of zone j changes by a unit, the required adjustment of the network flow rate (cubic meters per hour) for zone i to maintain the hydraulic balance of the entire network. For example, the calculated element a_12 = 0.05 means that if the heat load demand of zone 2 increases by 1 degree Celsius per hour, the flow rate of the corresponding branch in zone 1 needs to increase by 0.05 cubic meters per hour.

[0166] When executing S4-3, the alternating direction multiplier method is selected as the distributed optimization algorithm. The global hydraulic and thermal balance optimization problem is decomposed into three local subproblems of the main pipeline branches. Each subproblem solves its own flow allocation scheme in parallel. In each iteration, the coupling sensitivity matrix is ​​used to predict the feasibility of the flow scheme of each subproblem. For example, it is predicted whether the flow scheme of branch A will lead to insufficient pressure in branch B. If a potential imbalance risk is found, the flow allocation direction is adjusted according to the degree of impact quantified by the matrix elements. For example, the flow of branch A is reduced by 0.3 cubic meters per hour, and the flow of branch B is increased by 0.2 cubic meters per hour simultaneously to correct the scheme.

[0167] When executing S4-4, the preset convergence accuracy is 0.5%, meaning that convergence is determined when the differences in parameters such as flow rate and heat source output of each branch are less than 0.5% in two consecutive iterations. After 12 iterations, the algorithm meets the convergence condition and outputs a coordinated control instruction set: heat source 1 output 60 degrees Celsius per hour, heat source 2 output 70 degrees Celsius per hour; main pipeline A valve opening 75%, branch B valve opening 80%, branch C valve opening 65%; the operating frequencies of the three circulating pumps are 48 Hz, 46 Hz, and 42 Hz, respectively. The instruction set ensures the hydraulic and thermal balance of the pipeline network and meets the heat load requirements of all zones.

[0168] Please see Figure 2 It illustrates a schematic diagram of the system configuration of an intelligent scheduling and optimization system for energy-saving heating provided in an embodiment of the present invention, including:

[0169] The central processing platform is used to calculate and publish the marginal energy price of the system, receive the room temperature setpoint decision results of each heating zone, and generate a set of coordinated control instructions based on the decision results of all zones.

[0170] Multiple zone controllers are deployed in each heating zone to store and call the adjustment cost function of their respective zone, receive the marginal energy price of the system, and execute comparison logic to output the final decision setpoint of the zone.

[0171] The communication network connects the central processing platform with all zone controllers and is used to transmit the system's marginal energy price, the final decision setpoints for each zone, and the collaborative control instruction set.

[0172] It should be noted that this solution overcomes the limitations of traditional centralized scheduling (slow response and lack of personalization) and distributed scheduling (lack of global coordination) through its "centralized coordination + regional autonomy" architecture. Specifically, firstly, the central processing platform calculates a globally unified marginal energy price for the system using an iterative convergence algorithm, providing a consistent economic decision-making benchmark for all zones. This avoids energy waste across the entire network due to local optimization, ensuring both the globality and economy of the scheduling. Simultaneously, each zone controller locally stores its own adjustment cost function. Upon receiving price signals, it independently executes comparison logic and quickly outputs personalized room temperature setpoints, eliminating reliance on centralized decision-making by the central processing platform. This significantly improves response speed and adapts to the different building characteristics and user needs of various zones. Secondly, the communication network ensures real-time data interaction between the central platform and the zones, enabling the central processing platform to accurately aggregate network-wide demands. Through optimization algorithms, it generates a collaborative control instruction set that satisfies the hydraulic and thermal balance, ensuring that the needs of each zone are met while achieving optimal resource allocation across the entire network. Finally, the entire system forms a complete closed loop of "global pricing - regional decision-making - network-wide collaboration", which not only leverages the global optimization advantages of centralized scheduling, but also has the flexible response capabilities of distributed scheduling, effectively balancing economy, comfort and system stability, and resolving the contradictions between global and local, efficiency and accuracy in traditional scheduling.

[0173] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for intelligent dispatch optimization for energy-saving heat supply, characterized in that, The method comprises the following steps: Step S1, for each heating sub-zone in the heating network, a regulation cost function is defined, the heating sub-zone is a relatively independent control unit divided according to building type, insulation level or geographical location, the regulation cost function maps the deviation of the user room temperature set value of the sub-zone from the historical statistical reference value to the marginal cost change required for the heating system to achieve the deviation, the historical statistical reference value is the room temperature mean value calculated by excluding outliers and weighted averaging based on the normal operation data of the sub-zone in the past at least one heating season; Step S2, at each scheduling time, the system marginal energy price representing the global energy consumption cost of the system is obtained; Step S3, for each heating sub-zone, the system marginal energy price is compared with the current regulation cost function of the sub-zone, and the final decision set value of the sub-zone is independently decided according to the comparison result; Step S4, according to the final decision set value of all heating sub-zones, the collaborative control instruction set of heat source output, pipe network valve group and circulating pump that satisfies the hydraulic and thermal balance constraints of the heating network is solved.

2. The intelligent scheduling optimization method for energy-saving heating according to claim 1, characterized in that: wherein in step S1, the following sub-steps are further included: S1-1, for each heating sub-zone in the heating network, historical operation data of the sub-zone is collected, the operation data includes indoor and outdoor temperature, valve opening, supply and return water temperature, user room temperature set value change and corresponding system total energy consumption change; S1-2, the mapping relationship between the input of user room temperature set value change and the output of marginal cost converted from the corresponding system total energy consumption change is fitted through a machine learning algorithm; S1-3, the fitted mapping relationship is subjected to monotonicity constraint processing to ensure that the mapping relationship is a monotonically increasing function; S1-4, the processed monotonically increasing function is subjected to normalization and smoothing processing to obtain the regulation cost function of the heating sub-zone. 3.The energy-saving heat supply oriented intelligent dispatching optimization method according to claim 2, characterized in that, In the sub-step S1-3, the step of ensuring that the mapping relationship is a monotonically increasing function comprises: In the model training stage of the machine learning algorithm, a monotonicity regularization penalty term is added to the loss function, the monotonicity regularization penalty term L_mono is calculated according to the following formula: L_mono=λ*Σ_imax(0,-∇f(x_i)), wherein λ is a regularization intensity coefficient, Σ_i represents summing over the training sample index i, f(x_i) is the predicted output of the model at input x_i; ∇f(x_i) is the gradient of the model predicted output f at x_i with respect to the input x, the max(0,-∇f(x_i)) function ensures that only the negative gradient is penalized; By optimizing the loss function with the monotonicity regularization penalty term, the mapping relationship obtained by training keeps the output value monotonically increasing with the input value within its domain.

4. The intelligent scheduling optimization method for energy-saving heating according to claim 1, characterized in that: wherein in step S2, the following sub-steps are further included: S2-1, retrieving the historical marginal price under similar working conditions from the pre-trained price history pattern library according to the predicted outdoor temperature at the next scheduling time and the time period to which the predicted outdoor temperature belongs, and taking the historical marginal price as an initial system marginal energy price prediction value; S2-2, based on the current system marginal energy price prediction value and the regulation cost function of each heating subzone, simulating and predicting the room temperature set value adjustment decision of each heating subzone at the next scheduling time in parallel, and obtaining a predicted system total heat load by summarizing; S2-3, according to the predicted system total heat load and the current operating condition, combining the average temperature rise-cost mapping relationship fitted based on system historical data, calculating the marginal cost corresponding to the load change to serve as a candidate marginal energy price of this round; S2-4, calculating the difference between the candidate marginal energy price and the current system marginal energy price prediction value, and if the difference is less than a preset convergence threshold, publishing the candidate marginal energy price as the final system marginal energy price; S2-5, if the difference is greater than or equal to the preset convergence threshold, taking the candidate marginal energy price as a new round of system marginal energy price prediction value, and returning to step S2-2 for iteration.

5. The intelligent scheduling optimization method for energy-saving heating according to claim 1, characterized in that: wherein in step S3, the following sub-steps are further included: S3-1, for the currently processed heating subzone, solving a theoretical optimal room temperature set value according to its regulation cost function, so that under the set value, the marginal cost represented by the regulation cost function is equal to the system marginal energy price; S3-2, based on the user thermal comfort contract and the current indoor and outdoor state, obtaining the allowed room temperature set value range of the heating subzone; S3-3, according to a pre-defined constraint projection rule, mapping the theoretical optimal room temperature set value to the allowed room temperature set value range to obtain a final decision set value. 6.The energy-saving oriented intelligent dispatching optimization method for heating supply according to claim 5, characterized in that, In the sub-step S3-1, the solving of the theoretical optimal room temperature set value is realized by the following way: obtaining the historical statistical base room temperature T_base of the current heating subzone; when the regulation cost function is in an analytical form, solving the equation f(x)=P to obtain an optimal deviation amplitude x_opt, and calculating the theoretical optimal room temperature set value T_opt through the formula T_opt=T_base+x_opt, wherein f(x) is the regulation cost function, and P is the system marginal energy price; when the regulation cost function is a discrete lookup table or a black box model, a bisection method or a golden section method is used to find an x value in a preset interval, so that the absolute value of the difference between f(x) and P is less than a preset precision threshold, and the x value is taken as the theoretical optimal room temperature set value. 7.The energy-saving oriented smart dispatching optimization method for heating supply according to claim 6, characterized in that, In the sub-step S3-2, the allowed room temperature set value range is jointly defined by a lower limit value T_min and an upper limit value T_max. The lower limit value T_min is dynamically determined according to the following rule: T_min = max (T_contract, T_outdoor + ΔT_safe), wherein T_contract is an absolute minimum temperature specified in the user thermal comfort contract, T_outdoor is the current outdoor temperature, and ΔT_safe is a safety temperature difference for preventing pipe freezing, which is determined according to the building insulation performance, and ΔT_safe is determined according to the insulation performance level, and the higher the insulation performance level, the smaller the value of ΔT_safe, and the value range is 5-8℃; The upper limit value T_max is a preset fixed constant or a fixed value specified in the user thermal comfort contract. 8.The energy-saving oriented smart dispatching optimization method for heating supply according to claim 7, characterized in that, In the sub-step S3-3, the predefined constraint projection rule is: Compare the theoretical optimal room temperature setting value x_opt with the lower limit value T_min and the upper limit value T_max of the allowed room temperature setting value range; If x_opt < T_min, set the final decision setting value as the lower limit value; If x_opt > T_max, set the final decision setting value as T_max; If T_min ≤ x_opt ≤ T_max, set the final decision setting value as the theoretical optimal room temperature setting value.

9. The intelligent scheduling optimization method for energy-saving heating according to claim 1, characterized in that: In step S4, the following sub-steps are further included: S4-1, determine the full-network heat load distribution by taking the final decision setting values of all heating sub-areas as the heat load demands that must be met; S4-2, calculate the first-order derivative of the regulation cost function of each heating sub-area at its final decision setting value as the internal sensitivity of the heat demand of the sub-area to the change of its flow, combine the internal sensitivities of all sub-areas in the full network into a diagonal matrix, and multiply it by a constant matrix representing the hydraulic resistance of the pipe network to obtain a coupling sensitivity matrix; S4-3, use a distributed optimization algorithm to iteratively solve the collaborative control instructions that meet the hydraulic and thermal balance, and in each iteration, use the coupling sensitivity matrix to perform feasibility prediction and direction correction on the calculated flow distribution scheme; S4-4, when the iterative solution converges to a preset precision, output the set of collaborative control instructions.

10. A smart dispatch optimization system for energy saving heating, for implementing the method of any one of claims 1 to 9, characterized in that, It includes: A central processing platform for calculating and publishing the system marginal energy price, receiving the room temperature setting value decision results of each heating sub-area, and solving and generating a set of collaborative control instructions according to the decision results of all sub-areas; A plurality of sub-area controllers respectively deployed in each heating sub-area for storing and calling the regulation cost function of the sub-area, receiving the system marginal energy price, and executing comparison logic to output the final decision setting value of the sub-area; A communication network connecting the central processing platform and all sub-area controllers for transmitting the system marginal energy price, the final decision setting value of each sub-area, and the set of collaborative control instructions.

Images