Group control optimization method and device for air source heat pump unit group

Abstract

This invention provides a group control optimization method and device for air source heat pump unit groups. It is applied to the fields of regional heat source systems and energy-saving control technology. For the current rolling optimization cycle with N consecutive preset time steps, the following steps are executed: Inputting historical heating load value sequences, historical environmental parameter value sequences, and estimated environmental parameter value sequences into a heating load prediction model, outputting a predicted heating load value sequence; constructing a mixed integer nonlinear programming model based on the predicted heating load value sequence, the operating parameters of each unit, and the installation distance between every two units; solving the mixed integer nonlinear programming model to determine the unit start-stop combinations and the load rate values ​​allocated to each unit for each preset time step; adjusting the load rate values ​​of each unit to obtain the target load rate values ​​for each unit; controlling the air source heat pump unit group from the current moment until the predetermined time step ends, entering the next rolling optimization cycle.

Description

Technical Field

[0001] This invention relates to the field of regional heat source systems and energy-saving control technology, specifically to a group control optimization method and device for air source heat pump unit groups. Background Technology

[0002] In district heating systems, air source heat pump units, as a highly efficient and energy-saving heating device, have been widely used. However, with the expansion of heating system scale and the increasing complexity of operating conditions, traditional air source heat pump unit group control methods have gradually revealed many shortcomings. For example, air source heat pump unit groups suffer from problems such as localized heat accumulation, uncertainties in load forecasting, and difficulties in solving mixed-variable optimization problems during heating operation. Summary of the Invention

[0003] In view of this, the present invention provides a group control optimization method and apparatus for air source heat pump unit groups.

[0004] One aspect of the present invention provides a group control optimization method for an air source heat pump unit group, comprising: for a current rolling optimization cycle with N consecutive preset time steps starting from the current time, performing the following steps, where N is an integer greater than or equal to 3: inputting the historical heating load value sequence, the historical environmental parameter value sequence, and the estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group into a heating load prediction model, and outputting a predicted heating load value sequence within the current rolling optimization cycle; based on the predicted heating load value sequence, the operating parameters of each unit in the air source heat pump unit group, and the installation distance between every two units, constructing a hybrid... The total operating cost of the integer nonlinear programming model includes energy consumption cost, unit start-up and shutdown switching cost, and spatial layout cost among simultaneously operating units. The mixed integer nonlinear programming model is solved to determine the unit start-up / shutdown combinations and the assigned load rate values ​​for each unit at each preset time step within a decision space composed of unit start-up / shutdown state variables and discretized load rate variables. Furthermore, with fixed unit start-up / shutdown combinations, the load rate values ​​for each unit are adjusted to obtain the target load rate values ​​for each unit. Based on the unit start-up / shutdown combinations and the target load rate values ​​for each unit at a predetermined time step starting from the current moment, the air source heat pump unit group is controlled from the current moment until the predetermined time step ends, entering the next rolling optimization cycle.

[0005] According to an embodiment of the present invention, the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of an air source heat pump unit group are input into a heating load prediction model, and the predicted heating load value sequence within the current rolling optimization cycle is output. This includes: aligning the historical heating load value sequence and the historical environmental parameter value sequence with the time periods within the historical time periods, and combining them to obtain a first matrix; aligning the estimated environmental parameter value sequence with the time periods within the current rolling optimization cycle, and combining them to obtain a second matrix, where the heating load value is empty; concatenating the first matrix and the second matrix in chronological order to obtain an input matrix; performing embedding mapping and position encoding on the input matrix to obtain a temporal embedding representation; using a convolutional neural network to extract local features representing fluctuation and periodic patterns from the temporal embedding representation; using a multi-head self-attention mechanism to determine dependencies from the temporal embedding representation to obtain global features; fusing the local features and global features to obtain fused features, and mapping the features in the fused features corresponding to the current rolling optimization cycle to obtain the predicted heating load value sequence within the current rolling optimization cycle.

[0006] According to an embodiment of the present invention, the group control optimization method further includes: constructing constraints for a mixed integer nonlinear programming model; wherein the constraints include: robust load balance constraints; the robust load balance constraints are configured to: ensure that the real-time total heat supply of the operating units is not less than the sum of the predicted heat load value at the corresponding time in the predicted heat load value sequence and a first safety margin based on confidence level; ensure that the maximum total heating capacity of the operating units is not less than the sum of the predicted heat load value at the corresponding time in the predicted heat load value sequence and a second safety margin based on the statistical characteristics of prediction error, wherein the second safety margin is greater than the first safety margin.

[0007] According to an embodiment of the present invention, the constraints further include a minimum operating time constraint and a minimum downtime constraint; wherein, the minimum operating time constraint is configured to limit the minimum duration of continuous operation after the unit starts; and the minimum downtime constraint is configured to limit the minimum duration of continuous downtime after the unit stops.

[0008] According to an embodiment of the present invention, the constraints further include at least one of the following: operating state constraints, load rate constraints, and switching frequency constraints; the operating state constraints are configured to constrain the operating state of each unit based on the rated heating capacity and the predicted total heating load of each unit, wherein the predicted total heating load is the sum of each predicted heating load value in the predicted heating load value sequence; the load rate constraints are configured to constrain the load rate of each unit based on the load rate value range of each unit; and the switching frequency constraints are configured to constrain the switching frequency of the operating state of each unit based on a switching frequency threshold within the current rolling optimization cycle.

[0009] According to an embodiment of the present invention, the load rate range includes an upper limit and a lower limit. For each unit, the upper limit of the load rate is determined according to the following steps: based on the unit's operating parameters and rated power parameters, a power model is constructed in the form of a quadratic function; based on the relationship between the heating power and the load rate determined by the power model, the load rate value is determined when the difference between the unit's power and the power setpoint satisfies a preset condition; and the load rate value is determined as the upper limit of the load rate.

[0010] According to an embodiment of the present invention, the group control optimization method further includes: determining the energy consumption cost based on the heating and power output of each unit and the start-up and shutdown status of each unit, wherein the heating and power output is calculated based on the power model and operating parameters; calculating the unit start-up and shutdown switching cost based on the change in the start-up and shutdown status of each unit at adjacent times; calculating the spatial layout cost based on the reciprocal of the installation distance between any two units operating simultaneously; and determining the total operating cost by summing the energy consumption cost, the unit start-up and shutdown switching cost, and the spatial layout cost.

[0011] According to an embodiment of the present invention, the decision space includes multiple candidate solutions, each candidate solution indicating a candidate unit start-up / shutdown combination for each unit within a preset time step, and a candidate load factor value pre-allocated to each unit corresponding to the candidate unit start-up / shutdown combination. In the decision space composed of unit start-up / shutdown state variables and discretized load factor variables, determining the unit start-up / shutdown combination for each preset time step and the load factor value allocated to each unit includes: using a discrete particle swarm optimization algorithm, iteratively determining a target candidate solution from among the multiple candidate solutions in the decision space based on the constraints of a mixed-integer nonlinear programming model; determining the candidate unit start-up / shutdown combination within each preset time step indicated by the target candidate solution as the unit start-up / shutdown combination for the corresponding preset time step; and determining the candidate load factor value corresponding to the candidate unit start-up / shutdown combination as the load factor value.

[0012] According to an embodiment of the present invention, in the case of a fixed unit start-stop combination, adjusting the load rate value of each unit to obtain the target load rate value of each unit includes: based on the load rate value of each unit, treating the load rate variable as a continuous variable, determining the candidate value range of each load rate value; using a gradient descent algorithm to iteratively adjust the load rate value of each unit, wherein in each iteration, the gradient of the total operating cost with respect to the load rate variable is calculated, and the load rate value of each unit is updated along the descent direction that satisfies the constraint conditions; when it is determined that the convergence condition is met, the iteration is stopped, and the updated load rate value of each unit is determined to be the corresponding target load rate value.

[0013] Another aspect of the present invention provides a group control optimization device for an air source heat pump unit group, comprising: for a current rolling optimization cycle starting from the current time and spaced at N consecutive preset time steps, executing the following modules, where N is an integer greater than or equal to 3: a prediction module, used to input the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group into a heating load prediction model, and output the predicted heating load value sequence within the current rolling optimization cycle; and a construction module, used to construct a hybrid system based on the predicted heating load value sequence, the operating parameters of each unit in the air source heat pump unit group, and the installation distance between every two units, with the optimization objective of minimizing the total operating cost within the current rolling optimization cycle. The mixed-integer nonlinear programming model has a total operating cost that includes energy consumption cost, unit start-up and shutdown switching cost, and spatial layout cost among simultaneously operating units. The solution module solves the mixed-integer nonlinear programming model to determine the unit start-up / shutdown combinations and the assigned load rate values ​​for each unit at each preset time step within a decision space composed of unit start-up / shutdown state variables and discretized load rate variables. It also adjusts the load rate values ​​of each unit under a fixed start-up / shutdown combination to obtain the target load rate value for each unit. The control module controls the air-source heat pump unit group from the current moment, based on the unit start-up / shutdown combinations and the target load rate values ​​of each unit at a predetermined time step starting from the current moment, until the predetermined time step ends and the next rolling optimization cycle begins.

[0014] According to an embodiment of the present invention, by employing a rolling optimization strategy, after performing optimization control for a preset time step in each consecutive rolling optimization cycle of multiple preset time steps, the heating load is re-predicted, and the unit start-up / shutdown combinations and the load rate values ​​allocated to each unit for each preset time step within the rolling optimization cycle are re-solved based on the latest prediction results. This allows for the execution of optimization control for the next preset time step, thereby achieving dynamic adaptive adjustment and continuous optimized operation of the air source heat pump unit group. Furthermore, since the spatial layout cost between operating units is introduced into the total operating cost, the spatial layout can be optimized to avoid heat accumulation. However, directly incorporating this nonlinear, highly coupled physical constraint of spatial layout cost into a complex mixed-integer nonlinear programming model drastically increases the difficulty of solving the model, potentially leading to the inability to obtain a feasible solution within the preset time step of the rolling optimization cycle. Therefore, by utilizing the decision space composed of unit start-stop state variables and discretized load rate variables, the start-stop combinations of units at each preset time step and the load rate values ​​allocated to each unit are first solved. Then, the start-stop combinations of units are fixed, and the load rate values ​​of each unit are finely adjusted. Thus, while ensuring solution efficiency, the complex optimization objective, which includes energy consumption costs, unit start-stop switching costs, and spatial layout costs between simultaneously operating units, is transformed into high-precision control commands that can be practically solved and executed, thereby improving the control accuracy of the air source heat pump unit group. Attached Figure Description

[0015] The above and other objects, features and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings.

[0016] Figure 1 A flowchart of a group control optimization method for a group of air source heat pump units according to an embodiment of the present invention is shown.

[0017] Figure 2 A schematic diagram illustrating the determination of unit start-up and shutdown combinations for each preset time step and the load rate values ​​allocated to each unit according to an embodiment of the present invention is shown.

[0018] Figure 3 A schematic diagram illustrating the adjustment of load rate values ​​for each unit according to an embodiment of the present invention is shown.

[0019] Figure 4 A schematic diagram illustrating the determination of the predicted heating load value sequence within the current rolling optimization cycle according to an embodiment of the present invention is shown.

[0020] Figure 5 A block diagram of a group control optimization device for a group of air source heat pump units according to an embodiment of the present invention is shown. Detailed Implementation

[0021] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0022] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0023] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0024] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0025] In realizing the concept of this invention, it was discovered that the operating efficiency of air source heat pump unit groups is highly dependent on the accurate prediction of heating load. However, using simple linear models or statistical methods based on historical data results in low accuracy in load prediction, which cannot accurately reflect changes in actual heating demand. This can lead to situations where the unit provides excessive or insufficient heating during operation, wasting energy and potentially affecting the stability of the heating system and user comfort.

[0026] Furthermore, the maximum heating capacity of an air source heat pump unit is affected by various factors, including ambient temperature, unit operating status, and frost accumulation. However, assessments of the unit's maximum heating capacity are usually based on theoretical values ​​or data under nominal operating conditions, neglecting the performance degradation and dynamic changes of the air source heat pump unit in actual operation. This results in the unit being unable to meet heating requirements under high load demand.

[0027] Furthermore, the operating strategy of air source heat pump unit groups typically determines the start-up, shutdown, and operating status of the units based solely on preset temperature thresholds or simple load demands. However, this strategy is simplistic and fails to adequately consider factors such as the spatial spacing between units, power consumption, penalties for frequent start-ups and shutdowns, and cumulative operating time. For example, unreasonable spacing between active units can lead to localized cold accumulation or uneven cold distribution, affecting heating efficiency. Unoptimized unit load rates result in high total power consumption for the unit group, increasing operating costs. Frequent start-ups and shutdowns increase energy consumption and unit wear, shortening their lifespan. Overuse of some units while others remain idle leads to uneven wear between units, impacting system reliability and stability.

[0028] Based on this, embodiments of the present invention provide a group control optimization method for air source heat pump unit groups, including: for a current rolling optimization cycle with N consecutive preset time steps starting from the current time, performing the following steps, where N is an integer greater than or equal to 3: inputting the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group into a heating load prediction model, and outputting a predicted heating load value sequence within the current rolling optimization cycle; based on the predicted heating load value sequence, the operating parameters of each unit in the air source heat pump unit group, and the installation distance between every two units, constructing a system with the goal of minimizing the total operating cost within the current rolling optimization cycle. The mixed-integer nonlinear programming model has a total operating cost that includes energy consumption cost, unit start-up and shutdown switching cost, and spatial layout cost among simultaneously operating units. The mixed-integer nonlinear programming model is solved to determine the unit start-up / shutdown combinations and the assigned load rate values ​​for each unit at each preset time step within a decision space composed of unit start-up / shutdown state variables and discretized load rate variables. Furthermore, with fixed unit start-up / shutdown combinations, the load rate values ​​for each unit are adjusted to obtain the target load rate values ​​for each unit. Based on the unit start-up / shutdown combinations and the target load rate values ​​for each unit at a predetermined time step starting from the current moment, the air source heat pump unit group is controlled from the current moment until the predetermined time step ends, entering the next rolling optimization cycle.

[0029] The following will be through Figures 1-4 The group control optimization method for air source heat pump unit groups according to embodiments of the present invention will be described in detail.

[0030] Figure 1 A flowchart of a group control optimization method for a group of air source heat pump units according to an embodiment of the present invention is shown.

[0031] like Figure 1As shown, the method includes performing the following steps S110 to S140 for the current rolling optimization cycle, which starts from the current time and is spaced N consecutively at preset time steps. N is an integer greater than or equal to 3.

[0032] In step S110, the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group are input into the heating load prediction model, and the predicted heating load value sequence within the current rolling optimization cycle is output.

[0033] In step S120, based on the predicted heating load value sequence, the operating parameters of each unit in the air source heat pump unit group, and the installation distance between every two units, a mixed integer nonlinear programming model is constructed with the goal of minimizing the total operating cost within the current rolling optimization cycle.

[0034] In step S130, the mixed integer nonlinear programming model is solved to determine the unit start-stop combinations and the load rate values ​​allocated to each unit in the decision space composed of unit start-stop state variables and discretized load rate variables for each preset time step; and, under the condition of fixed unit start-stop combinations, the load rate values ​​of each unit are adjusted to obtain the target load rate values ​​of each unit.

[0035] In step S140, based on the unit start-stop combination with a predetermined time step starting from the current time and the target load rate value of each unit, the air source heat pump unit group is controlled from the current time until the predetermined time step ends and the next rolling optimization cycle begins.

[0036] In embodiments of the present invention, any air source heat pump unit within an air source heat pump unit group is simply referred to as a unit for ease of description. The historical heating load value sequence may include heating load values ​​corresponding to multiple historical moments. The historical environmental parameter value sequence may include parameter values ​​of environmental parameters corresponding to multiple historical moments. The estimated environmental parameter value sequence may include parameter values ​​of environmental parameters corresponding to multiple future moments. Environmental parameters may include, but are not limited to, outdoor temperature, outdoor humidity, and indoor temperature. Future moments are moments after the current moment, and historical moments are moments before the current moment.

[0037] The heating load prediction model can be a long short-term memory network, without specific limitations. For example, it can be based on aligning historical heating loads with historical environmental parameters over time to form a fixed-length historical window sequence. The estimated environmental parameter sequence within the current rolling optimization cycle is concatenated with the historical window sequence as model input. The long short-term memory network processes the input sequence and directly outputs the predicted heating load values ​​for each moment within the current rolling optimization cycle.

[0038] It should be noted that after each group control optimization is completed, the newly collected actual data can be added to the historical window sequence, and the model input can be updated in a rolling manner to predict the next rolling optimization cycle.

[0039] Total operating costs may include: energy consumption costs, unit start-up and shutdown switching costs, and space layout costs between units operating simultaneously.

[0040] The decision space can be composed of start / stop state variables and discretized load rate variables. Start / stop state variables can be 0 or 1. The discretized load rate variables can be multiple discrete values ​​with a step size of 0.1. This allows for a smaller search size when using genetic algorithms or particle swarm optimization to perform a global search within the decision space, avoiding early entrapment in local optima. Since the discretized load rate variables are limited by the step size, they may skip the true optimal load rate. Therefore, under a fixed start / stop combination, adjusting the load rate values ​​of each unit to obtain the target load rate value for each unit can avoid deviation from the true optimal load rate value, improving load rate control accuracy while accelerating the solution speed.

[0041] According to an embodiment of the present invention, by employing a rolling optimization strategy, after performing optimization control for a preset time step in each consecutive rolling optimization cycle of multiple preset time steps, the heating load is re-predicted, and the unit start-up / shutdown combinations and the load rate values ​​allocated to each unit for each preset time step within the rolling optimization cycle are re-solved based on the latest prediction results. This allows for the execution of optimization control for the next preset time step, thereby achieving dynamic adaptive adjustment and continuous optimized operation of the air source heat pump unit group. Furthermore, since the spatial layout cost between operating units is introduced into the total operating cost, the spatial layout can be optimized to avoid heat accumulation. However, directly incorporating this nonlinear, highly coupled physical constraint of spatial layout cost into a complex mixed-integer nonlinear programming model drastically increases the difficulty of solving the model, potentially leading to the inability to obtain a feasible solution within the preset time step of the rolling optimization cycle. Therefore, by utilizing the decision space composed of unit start-stop state variables and discretized load rate variables, the start-stop combinations of units at each preset time step and the load rate values ​​allocated to each unit are first solved. Then, the start-stop combinations of units are fixed, and the load rate values ​​of each unit are finely adjusted. Thus, while ensuring solution efficiency, the complex optimization objective, which includes energy consumption costs, unit start-stop switching costs, and spatial layout costs between simultaneously operating units, is transformed into high-precision control commands that can be practically solved and executed, thereby improving the control accuracy of the air source heat pump unit group.

[0042] According to an embodiment of the present invention, the group control optimization method may include, in addition to, the following: Figure 1In addition to steps S110 to S140, the following operations may also be included: determining the energy consumption cost based on the heating and electrical power of each unit and the start-up and shutdown status of each unit, wherein the heating and electrical power is calculated based on the power model and operating parameters; calculating the unit start-up and shutdown switching cost based on the change in the start-up and shutdown status of each unit at adjacent times; calculating the spatial layout cost based on the reciprocal of the installation distance between any two units operating simultaneously; and determining the total operating cost by summing the energy consumption cost, the unit start-up and shutdown switching cost, and the spatial layout cost.

[0043] For example, the objective function of a mixed-integer nonlinear programming model can be expressed as follows (1):

[0044] (1);

[0045] in, For the current rolling optimization cycle, For the current moment, This refers to the number of units within the air source heat pump unit group. For the first Taiwanese crew at the current moment The start / stop status is 0 or 1, where 0 indicates shutdown and 1 indicates operation. For the first Taiwanese crew at the current moment The load factor can be determined based on the ratio of the predicted heating load to the rated heating capacity. Outdoor ambient temperature under specific operating conditions This refers to the unit's outlet water temperature under specific operating conditions. For heating power; For the first Taiwan crew current time The heating power is expressed in kW. If the unit is shut down, the heating power is 0. The penalty coefficient for frequent unit start-ups and shutdowns is given by the current time t and the time at the current time. The previous moment The interval is preset time step. Indicates the first Taiwan crew current time Whether the operating status changes is used to describe whether the unit is frequently started and stopped. The penalty coefficient for insufficient installation distance between units; For the first Taiwan unit and the first The installation distance between the units is used to constrain the spatial arrangement requirements between the units; For the current moment , No. Start-up and shutdown status of the generator set; Indicates if the first Taiwan unit and the first The installation distance between two generating units when they are running simultaneously; This represents the sum of the installation distances between two units operating within an air source heat pump unit group, expressed in meters (m).

[0046] According to embodiments of the present invention, based on energy consumption cost, start-stop switching cost and space layout cost in the total operating cost, the actual power consumption can be accurately quantified while meeting heating demand, effectively suppressing frequent start-stop of units, and prioritizing the selection of units with decentralized layout to improve hydraulic and thermal distribution. The three factors work together to achieve unified optimization of energy saving, service life and system balance of air source heat pump unit groups.

[0047] According to an embodiment of the present invention, the decision space may include multiple candidate solutions, wherein the candidate solutions indicate the candidate unit start-up and shutdown combinations of each unit within each preset time step, and the candidate load rate values ​​pre-allocated to each unit corresponding to the candidate unit start-up and shutdown combinations.

[0048] Figure 2 A schematic diagram illustrating the determination of unit start-up and shutdown combinations for each preset time step and the load rate values ​​allocated to each unit according to an embodiment of the present invention is shown.

[0049] Within a decision space comprised of unit start-up and shutdown state variables and discretized load rate variables, determining the unit start-up and shutdown combinations for each preset time step and the load rate values ​​allocated to each unit can include, for example: Figure 2 Steps S210 to S230 are shown.

[0050] In step S210, the discrete particle swarm optimization algorithm is used to iteratively determine the target candidate solution from multiple candidate solutions in the decision space based on the constraints of the mixed integer nonlinear programming model.

[0051] In step S220, the candidate unit start-stop combinations within each preset time step indicated by the target candidate solution are determined as the unit start-stop combinations for the corresponding preset time step.

[0052] In step S230, the candidate load rate value corresponding to the candidate unit start-stop combination is determined as the load rate value.

[0053] During the iteration process, if there are start / stop states and load rates that meet the constraints and can minimize the value of the optimization objective, the start / stop states and load rates determined within the preset time steps other than the first preset time step in the current rolling optimization cycle can be used as the initial start / stop states and initial load rates for the next rolling optimization cycle, thereby reducing the program's computation time.

[0054] For example, if the current rolling optimization cycle has three preset time steps, after obtaining the target load rate values ​​and start-stop combinations for each unit within the first, second, and third preset time steps, control commands for each unit are generated based solely on the target load rate values ​​and start-stop combinations within the first preset time step, and control is applied to each unit for the first preset time step. Control for the second preset time step of each unit needs to be determined in the next rolling optimization cycle. For the second, third, and fourth preset time steps in the next rolling optimization cycle, the target load rate values ​​and start-stop combinations for each unit within the second and third preset time steps of the current rolling optimization cycle can be used as the initial start-stop states and initial load rates for the second and third preset time steps in the next rolling optimization cycle.

[0055] Since the accuracy of the load factor values ​​allocated to each unit in the first stage was only at the level of 0.1, further adjustments and optimizations can be made to bring the accuracy of the load factor allocation to the level of 0.01, thereby guiding actual production operations.

[0056] Figure 3 A schematic diagram illustrating the adjustment of load rate values ​​for each unit according to an embodiment of the present invention is shown.

[0057] According to an embodiment of the present invention, in the case of a fixed unit start-stop combination, adjusting the load rate value of each unit to obtain the target load rate value of each unit may include, for example: Figure 3 Steps S310 to S330 are shown.

[0058] In step S310, based on the load rate values ​​of each unit, the load rate variable is treated as a continuous variable, and the candidate value range for each load rate value is determined.

[0059] In step S320, the load rate values ​​of each unit are iteratively adjusted using the gradient descent algorithm.

[0060] In step S330, if the convergence condition is met, the iteration is stopped, and the updated load rate values ​​of each unit are determined to correspond to the target load rate values.

[0061] In each iteration, the gradient of total operating cost with respect to the load factor variable is calculated, and the load factor values ​​of each unit are updated in the descent direction that satisfies the constraints.

[0062] The start-stop combinations of each unit with a preset time step within the air source heat pump unit group can be used to... The combination of load factor values ​​allocated to each unit can be: , For the first Taiwanese crew at the current moment Maximum load rate.

[0063] fixed Only adjust This invention modifies the gradient descent algorithm used to a conventional algorithm, and the invention does not impose specific limitations. Convergence conditions may include at least one of the following: the change in total running cost is less than a predetermined change threshold, the gradient norm is less than a predetermined norm threshold, the number of iterations reaches the maximum number of iterations, etc. If higher accuracy is required, the convergence conditions can be adjusted.

[0064] According to an embodiment of the present invention, by relaxing the load factor from a discrete variable to a continuous variable and employing gradient descent for projection iteration, a high-precision optimal solution satisfying the convergence condition can be quickly converged under a fixed start-stop combination. This effectively compensates for the low accuracy problem caused by the discretization step size. This two-stage solution strategy improves the load factor control accuracy while ensuring solution efficiency, meeting the practical requirements of engineering for economy, comfort, and stability. Compared to directly solving mixed-integer nonlinear programming, this embodiment has a smaller computational load and is suitable for real-time operation within a rolling optimization framework embedded in model predictive control.

[0065] Figure 4 A schematic diagram illustrating the determination of the predicted heating load value sequence within the current rolling optimization cycle according to an embodiment of the present invention is shown.

[0066] According to an embodiment of the present invention, for step S110, the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group are input into the heating load prediction model, and the predicted heating load value sequence within the current rolling optimization cycle is output, which may include, for example... Figure 4 Steps S410 to S470 are shown.

[0067] In step S410, the historical heating load value sequence and the historical environmental parameter value sequence are aligned with the time periods within the historical time period and combined to obtain the first matrix.

[0068] In step S420, the estimated environmental parameter value sequence is aligned with the time within the current rolling optimization cycle and combined to obtain a second matrix, in which the heating load value is empty.

[0069] In step S430, the first matrix and the second matrix are concatenated in chronological order to obtain the input matrix.

[0070] In step S440, the input matrix is ​​subjected to embedding mapping and position encoding to obtain a temporal embedding representation.

[0071] In step S450, a convolutional neural network is used to extract local features from the temporal embedding representation to characterize fluctuation and periodic patterns.

[0072] In step S460, the multi-head self-attention mechanism is used to determine the dependencies from the temporal embedding representation and obtain the global features.

[0073] In step S470, local features and global features are fused to obtain fused features, and the features in the fused features that correspond to the current rolling optimization cycle are mapped to obtain the predicted heating load value sequence within the current rolling optimization cycle.

[0074] In this embodiment, the heating load prediction model can be pre-trained. The heating load prediction model may include, for example, an embedding layer, a convolutional layer of a convolutional neural network, a global dependency module of a Transformer, a feature fusion module, and a prediction layer.

[0075] The historical time period includes a first historical time period and a second historical time period. The second historical time period follows the first historical time period. The historical heating load value sequence can include historical heating loads corresponding to multiple historical moments. The historical environmental parameter value series can include outdoor temperatures corresponding to multiple historical moments. Outdoor humidity Indoor temperature wait.

[0076] As a preferred approach, before alignment, the historical heating load value sequence and the historical environmental parameter value sequence can be cleaned and preprocessed to remove outliers and missing values, and the values ​​in the two sequences can be standardized to eliminate the dimensional differences between different features and improve the prediction performance of the model.

[0077] Historical heating load The calculation method can be shown in the following formula (2):

[0078] (2);

[0079] in, The hot water supply temperature of the main pipe is expressed in °C. The temperature of the hot water return from the main pipe is in °C. This refers to the density of hot water, in units of... ; This refers to the specific heat capacity of hot water, in units of... ; The hot water flow rate of the main pipe is expressed in units of... ; This represents the heating load, expressed in kW.

[0080] Standardization can include Z-Score standardization, which can convert the values ​​in a sequence into a standard normal distribution with a mean of 0 and a standard deviation of 1, as shown in equation (3) below:

[0081] (3);

[0082] in, It can be any value from either of the two sequences, i.e., the original numerical value; The mean of the original values ​​in the two sequences; The standard deviation of the original values ​​in the two sequences; This is the standardized value, that is, the standardized value obtained by standardizing the original value.

[0083] The sample first matrix corresponding to the first historical time period and the sample second matrix corresponding to the second historical time period can be concatenated in chronological order to obtain the sample input matrix. The heating load value in the sample second matrix is ​​the sample label.

[0084] The pre-training method can include the following operations: The sample input matrix is ​​input into the embedding layer and mapped to a high-dimensional embedding space through a linear transformation to obtain a sample embedding vector. Then, positional encoding is introduced to incorporate positional information into the sample embedding vector, resulting in a temporal embedding representation of the samples. This temporal embedding representation is input into a convolutional neural network to extract local features representing fluctuation and periodic patterns. The temporal embedding representation is then input into a Transformer global dependency module, which uses a multi-head self-attention mechanism to determine sample dependencies, obtaining global features. The local and global features are input into a feature fusion module, which outputs fused features. These fused features are then input into a prediction layer to predict the heating load value sequence for a second historical time period, resulting in a predicted heating load value sequence for that period. Based on the predicted heating load value sequence and the historical heating load value sequence for the second historical time period, the heating load prediction model is trained, resulting in a trained heating load prediction model.

[0085] For example, the sample embedding vector can be represented as shown in equation (4):

[0086] (4);

[0087] in, To embed the weight matrix, the dimension is , For each time point, there are 4 feature dimensions, and the features can include heating load. Outdoor temperature Outdoor humidity Indoor temperature wait, Let the dimension of the target high-dimensional embedding space be 16. For bias terms, The sample embedding vector has a dimension of . , This represents the number of moments within a historical time period, i.e., the sequence length.

[0088] Position encoding can be based on sine and cosine functions, and no specific limitations are specified here. For example, samples can be embedded into vectors. and position encoding vector By adding them together, we obtain the temporal embedding representation of the samples. As shown in equation (5):

[0089] (5);

[0090] in, and Dimensions and They are the same, both are .

[0091] Temporal embedding representation of samples using convolutional layers of a convolutional neural network Perform convolution operations to extract local features of samples in the sequence. , For the first time period in history The temporal embedding representation of the samples at each time step is shown in Equation (6) below:

[0092] (6);

[0093] in, Indicates the first in a historical period At each time point, the local features of the corresponding sample are extracted through convolution operations. ; To utilize one-dimensional convolutional layers for the first Convolution operation is performed on the temporal embedding representation of the samples corresponding to each time step; The kernel size can be... Thus, the local features of the sample The dimension can be ; For the first One convolutional kernel vector; For the first Temporal embedding representation of samples corresponding to each time point; For the first Each convolution kernel bias term.

[0094] By utilizing the pooling layer of a convolutional neural network, the local features of the samples are further abstracted and dimensionality reduced to capture local patterns at different scales. This pooling operation can reduce the length of the sequence, improve the robustness of the features, and enhance the model's ability to extract local features, as shown in equation (7) below:

[0095] (7);

[0096] in, For the first Local features of samples extracted after pooling at each time step; This is the max pooling function, and the pooling kernel size can be 3.

[0097] Temporal embedding representation of samples with location information The input is fed into the global dependency module, which utilizes its multi-head self-attention mechanism to capture global dependencies between different time points in the sequence. Because the multi-head self-attention mechanism dynamically focuses on important time points and features in the sequence by calculating the similarity between queries, keys, and values, it is better able to model long-term dependencies. For the ... Each self-attention point is shown in equations (8) to (10):

[0098] (8);

[0099] (9);

[0100] (10);

[0101] in: For the first A query matrix, with dimensions of , The dimension of the multi-head self-attention mechanism can be 64; To query the weight matrix, the dimension is The parameter matrix can be learned; For the first A key matrix with dimension , ; The key weight matrix has dimensions of . The parameter matrix can be learned; For the first A matrix of values, with dimensions of . ; The value weight matrix has dimensions of . It can learn parameter matrices; query matrices Key matrix Value matrix .

[0102] Self-attention mechanism The calculation is shown in the following formula (11):

[0103] (11);

[0104] in, The softmax operation is used in machine learning to model sample dependencies; the multi-head self-attention mechanism calculates global features of the samples after passing through a linear layer. , dimension As shown in equation (12):

[0105] (12);

[0106] in, For query matrix Key matrix Value matrix Multi-headed attention; For splicing operations; The weight matrix for the multi-head self-attention mechanism concatenation operation has a dimension of . ; The bias is used for splicing operations in a multi-head self-attention mechanism.

[0107] Local features of the sample and global features of the sample The input feature fusion module performs fusion to fully utilize the advantages of both features. Fusion methods can include weighted summation, post-concatenation transformation, or gating mechanisms. The post-concatenation transformation concatenates the two features together and performs a non-linear transformation through a fully connected layer to obtain the sample fused features. As shown in equation (13):

[0108] (13);

[0109] in, To combine the local features and global features of a sample; It is a fully connected layer; , dimension , For the first The sample fusion features corresponding to each time point.

[0110] The fused features The input is fed into the prediction layer, which can be a fully connected layer, and outputs the predicted heating load value sequence for the second historical time period, as shown in equation (14) below:

[0111] (14);

[0112] in, Here is the prediction layer weight matrix, with dimension 1. ; For the bias of the prediction layer; For the second historical period Sample fusion features corresponding to each time point; For the first The predicted heating load value corresponding to each time point is expressed in kW.

[0113] The mean squared error loss function can be used to measure the loss between the predicted heating load value series and the historical heating load value series. As shown in equation (15), the model parameters are updated through the backpropagation algorithm to optimize the model performance.

[0114] (15);

[0115] in, The number of moments within the second historical time period; For the first The historical heating load value corresponding to each moment, in kW.

[0116] During training, optimizers or other optimization algorithms can be used, combined with strategies such as learning rate decay and early stopping, to prevent model overfitting and improve the model's generalization ability.

[0117] Embedding mapping can be used to map the input matrix to a high-dimensional embedding space through linear transformation, thereby increasing the expressive power of the data and obtaining an embedding vector. Then, positional encoding can be used to incorporate positional information into the embedding vector, resulting in a temporal embedding representation. Convolutional neural networks can be used to perform convolution operations on the temporal embedding representation to extract local features. The temporal embedding representation can be input into the Transformer's global dependency module, utilizing its multi-head self-attention mechanism to capture the dependencies between different time points in the sequence, and then the global features are calculated after passing through linear layers. The fused features can be input into a fully connected layer to output the predicted heating load value sequence within the current rolling optimization cycle.

[0118] According to an embodiment of the present invention, an input matrix including unknown load values ​​for future periods is formed by concatenating historical data and estimated environmental parameters for future periods, and location information is embedded in the input matrix. Based on this, convolutional layers of a convolutional neural network are used to accurately extract local features representing fluctuation and periodic patterns in the sequence, and a Transformer is used to model global dependencies across time periods to obtain global features. After fusing these two features, the features corresponding to the current rolling optimization cycle are mapped, which can effectively improve the accuracy of the predicted heating load value sequence within the current rolling optimization cycle.

[0119] According to embodiments of the present invention, the group control optimization method may include, in addition to, the following: Figure 1 In addition to steps S110 to S140, the operation may also include: constructing constraints for a mixed-integer nonlinear programming model.

[0120] Constraints may include robust load balance constraints. Robust load balance constraints are configured to ensure that the real-time total heating output of the operating units is not less than the sum of the predicted heating load value at the corresponding time in the predicted heating load value sequence and a first safety margin based on confidence level; and that the maximum total heating capacity of the operating units is not less than the sum of the predicted heating load value at the corresponding time in the predicted heating load value sequence and a second safety margin based on the statistical characteristics of prediction error, wherein the second safety margin is greater than the first safety margin.

[0121] In this embodiment of the invention, the robust load balance constraint is shown in equation (16):

[0122] (16);

[0123] in, The rated heat output of the unit is expressed in kW, which is the heating capacity under standard operating conditions. , for the first The ratio of the actual heating power of the unit to its rated heating capacity is an auxiliary variable. This refers to the actual heating power of the units operating within the air source heat pump group, expressed in kW. For the current moment... Heating load = , This represents the predicted heating load value at the current moment. For the current moment The standard deviation of the prediction error can be determined by constructing a load database, which can include a sequence of predicted heating load values ​​and a sequence of heating load values. When the sequence length accumulates to 24 or more, the standard deviation of the prediction error can be further determined by calculating the error of the predicted load value sequence. This indicates a 95% confidence level. This is the first safety margin based on confidence level; This is the unit's maximum heating capacity, measured in kW. It is usually greater than or equal to the rated heating capacity and reflects the unit's ultimate output. This is a second safety margin based on the statistical characteristics of prediction errors; This refers to the maximum total heating capacity of the units operating within a group of air source heat pump units.

[0124] Because load forecasting can be inaccurate, and the control commands are for the start / stop of air-source heat pump units and their outlet water temperature, if the predicted heating load is lower than expected, and the maximum heating capacity of the operating units exceeds or meets the actual load, only the return water temperature will decrease slightly, which can still basically meet the heating needs. However, if the maximum heating capacity of the operating units is lower than the actual load, not only will the return water temperature decrease, but the supply water temperature will also be difficult to reach the control value, resulting in unmet heating demand, lower indoor temperatures, and reduced heating comfort. Therefore, by setting robust load balance constraints with dual safety margins, the real-time total heating capacity of the operating units can meet the predicted load demand with a 95% confidence level. Simultaneously, the maximum total heating capacity of the units is required to reserve a larger standby capacity. This addresses load forecasting errors, ensures heating reliability, avoids excessive redundancy, and improves the economy and robustness of the heat pump group control system.

[0125] According to embodiments of the present invention, the constraints may further include a minimum operating time constraint and a minimum downtime constraint. The minimum operating time constraint is configured to limit the minimum duration of continuous operation after the unit starts up; the minimum downtime constraint is configured to limit the minimum duration of continuous downtime after the unit stops operating.

[0126] In this embodiment of the invention, the minimum running time constraint can be expressed linearly. For example, the start and stop flags are first defined as shown in the following equation (17):

[0127] (17);

[0128] in, For the first Taiwanese crew at the current moment The start-up flag; For the first Taiwanese crew at the current moment Stop sign; For the first Taiwanese crew at the current moment The previous moment The start and stop states are controlled to ensure that the start and stop states cannot both be 1 at the same time.

[0129] If the first Taiwanese crew at the current moment start up( If so, it must run continuously for at least For each time period, the minimum running time constraint is as shown in equation (18):

[0130] (18);

[0131] in, For the first Minimum operating time of the unit.

[0132] The minimum downtime constraint can be linearized if the i-th Taiwanese crew at the current moment Shut down ( If so, it must run continuously for at least For each time period, the minimum downtime constraint is as shown in equation (19):

[0133] (19);

[0134] in, For the first Minimum downtime for the unit.

[0135] Frequent start-ups and shutdowns are a major cause of accelerated wear and shortened lifespan in air source heat pump units. Therefore, this invention uses a minimum operating time constraint to accurately characterize the continuous operating duration of the unit, avoiding nonlinear logic judgments and enabling efficient solution of the mixed-integer nonlinear programming model. Furthermore, a minimum downtime constraint accurately characterizes the minimum continuous downtime of the unit. The combination of these two methods effectively prevents frequent start-ups and shutdowns, extends the lifespan of air source heat pump units, and ensures the solution stability of the mixed-integer nonlinear programming model, providing engineering feasibility for heat pump group control systems.

[0136] According to embodiments of the present invention, the constraints may further include at least one of the following: operating state constraints, load rate constraints, and switching frequency constraints.

[0137] Operating status constraints can be configured to constrain the operating status of each unit based on the rated heating capacity and the total heating load forecast. The total heating load forecast is the sum of each predicted heating load value in the predicted heating load value sequence.

[0138] The load factor constraint is configured to constrain the load factor of each unit based on the range of load factor values ​​for each unit.

[0139] The switching frequency constraint is configured to constrain the switching frequency of each unit's operating status based on a switching frequency threshold within the current rolling optimization cycle.

[0140] Because mixed-integer nonlinear programming models need to handle bilinearity... ,in It is a binary variable. It is a continuous variable. This mixed product term is nonlinear and nonconvex, and cannot be directly handled by the standard solver. Therefore, the behavior of the bilinear term at integer solutions can be accurately characterized by four linear inequalities, as shown in equation (20) below:

[0141] (20);

[0142] Therefore, the operating state constraints can constrain the start-up and shutdown states of each unit. It must be 0 or 1, as shown in equation (21) below:

[0143] (twenty one).

[0144] According to an embodiment of the present invention, the load factor range includes an upper limit and a lower limit. Load factor constraints can constrain the load factor of each unit. It must be between its upper and lower load factor limits, as shown in equation (22):

[0145] (twenty two).

[0146] Switching frequency constraints can limit the start-stop switching frequency of the unit to a certain range, avoiding excessive start-stop operations. The maximum start-stop frequency can be set. It satisfies the inequality shown in equation (23) below:

[0147] (twenty three);

[0148] Therefore, we can obtain the inequalities shown in equations (24) to (26) as follows:

[0149] (twenty four);

[0150] (25);

[0151] (26);

[0152] in, For the first Taiwanese crew at the current moment The start-up switching range.

[0153] According to embodiments of the present invention, by introducing operating state constraints, the units can be forced to maintain the corresponding state for at least a certain period of time after startup or shutdown, effectively curbing the damage to equipment life caused by frequent start-ups and shutdowns; load rate constraints control the output of each unit between the allowable minimum load rate and maximum load rate, avoiding inefficient operation or the risk of exceeding limits; switching frequency constraints further limit the number of times the units start up or stop or the load rate is adjusted per unit time, preventing drastic fluctuations in control commands. If these three constraints are combined, the operational stability, equipment reliability, and feasibility of optimization solutions of the heat pump group control system can be significantly improved while meeting heating demand.

[0154] According to an embodiment of the present invention, the load rate range includes an upper limit and a lower limit. For each unit, the upper limit of the load rate is determined according to the following steps: based on the unit's operating parameters and rated power parameters, a power model is constructed in the form of a quadratic function; based on the relationship between the heating power and the load rate determined by the power model, the load rate value when the difference between the unit's power and the power setpoint satisfies a preset condition is determined; and the load rate value is determined as the upper limit of the load rate.

[0155] Because quadratic functions are more flexible than linear models, they can capture nonlinear relationships in the data. Therefore, by introducing squared terms and cross terms, they can simulate complex interactions between variables. Using the recursive least squares algorithm, based on the unit's operating parameters and rated power parameters, a power model is constructed in the form of a quadratic function as shown in equation (27):

[0156] (27);

[0157] in, , , The heating capacity of the unit is expressed in kW. The rated heat supply of the unit, in kW; For unit load rate, Outdoor ambient temperature For the unit's water output, In order to be in The unit's electrical power; , Let represent a quadratic term, where the matrix represents a quadratic term. This corresponds to the quadratic coefficient matrix, the matrix The element in the i-th row and j-th column of the quadratic term is... The coefficients of the terms, i=1, 2, 3, j=1, 2, 3, middle , , ,matrix The element in the i-th row and j-th column is a diagonal element when i=j. (like The element in the first row and first column; The element in the 2nd row and 2nd column; (The element in the 3rd row and 3rd column); when i ≠ j, it is a non-diagonal element. and , (like The element in the 1st row and 2nd column; or The element in the 2nd row and 1st column; The element in the 1st row and 3rd column; or The element in the 3rd row and 1st column; The element in the 2nd row and 3rd column; or (element in row 3, column 2). Represents a linear term; , It is with variables The corresponding coefficients; The coefficient vector of the linear terms in the power model The row element; For the power model constants, The rated thermal power of the unit is expressed in kW.

[0158] The maximum operating load of an air-source heat pump unit under specific operating conditions can be calculated using the gradient descent method, and the outdoor ambient temperature under specific operating conditions can be obtained by looking up the performance parameter table using interpolation. Unit outlet water temperature under specific operating conditions Maximum heating capacity of the unit Corresponding operating power By employing the gradient descent method, the maximum heating capacity of each unit under different operating conditions can be accurately assessed, ensuring efficient operation of the units in various environmental conditions, thereby significantly improving the reliability of the heating system and user satisfaction.

[0159] Considering the performance degradation of the actual operating units, which leads to a decrease in their rated energy efficiency ratio, the unit load rate... Outdoor ambient temperature under specific operating conditions Unit outlet water temperature under specific operating conditions It can be considered a constant, the unit's electrical power Regarding unit load rate The gradient is calculated as shown in equation (28):

[0160] (28);

[0161] in, Electric power of the unit Regarding unit load rate The gradient.

[0162] Based on unit electrical power Regarding unit load rate gradient, unit load rate The update is shown in equation (29):

[0163] (29);

[0164] in, Updated load factor value; For learning factors.

[0165] The unit load rate can be continuously updated through iterative cycles. and unit power The iteration continues until a preset condition is met. The preset condition could be that the unit's electrical power... With power setting value The absolute value of the difference between the values ​​(design values ​​given by the equipment manufacturer under fixed temperature and humidity conditions) and the convergence criterion is less than the convergence criterion. ,Right now When the iteration stops, the upper limit of the load factor can be determined based on the load factor update value obtained after the iteration converges. The maximum heating capacity of the computer unit As shown in equation (30):

[0166] (30).

[0167] No. Taiwanese crew at the current moment The maximum load rate can be determined based on the upper limit of this load rate.

[0168] According to an embodiment of the present invention, by constructing a power model using a quadratic function, the nonlinear relationship between the unit's heating power and the load rate is accurately characterized. Then, based on this model, the load rate value when the difference between the power and the power setpoint meets the preset conditions is determined, and this value is used as the upper limit of the load rate. This can effectively constrain the unit's operating range, avoid the actual power deviating too much from the setpoint, thereby improving the accuracy of power control and system energy efficiency, while providing a reasonable and feasible load rate boundary for optimized scheduling.

[0169] According to embodiments of the present invention, through predictive and rolling optimization technologies, the operating status of the units can be automatically adjusted based on real-time monitoring data and predictive information, achieving intelligent and automated operation management. This intelligent control method reduces manual intervention, lowers the difficulty and cost of operation management, and improves the system's response speed and adaptability, enabling it to better cope with complex changes in heating conditions. Through intelligent control, the system can optimize operating strategies in real time, ensuring efficient operation under different conditions and further improving the overall performance of the district heat source system. Furthermore, through high-precision load forecasting, accurate assessment of the unit's maximum heating capacity, and optimization of unit group operating strategies, the present invention can significantly reduce the operating and maintenance costs of the heating system, improve energy efficiency, and extend equipment lifespan. This not only provides users with a more stable and comfortable heating service but also enhances the economic viability and competitiveness of the district heat source system in the market. In addition, the technological innovation of the present invention provides strong support for the widespread application and development of air source heat pump technology in the field of district heating, promotes technological progress in related industries, and has significant market application prospects and social value.

[0170] Figure 5 A block diagram of a group control optimization device for a group of air source heat pump units according to an embodiment of the present invention is shown.

[0171] like Figure 5 As shown, the group control optimization device 500 for air source heat pump unit group includes the following modules, where N is an integer greater than or equal to 3, for the current rolling optimization cycle starting from the current time and with N consecutive preset time steps: prediction module 510, construction module 520, solution module 530 and control module 540.

[0172] The prediction module 510 is used to input the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group into the heating load prediction model, and output the predicted heating load value sequence within the current rolling optimization cycle.

[0173] Module 520 is used to construct a mixed integer nonlinear programming model based on the predicted heating load value sequence, the operating parameters of each unit in the air source heat pump unit group, and the installation distance between every two units, with the optimization objective of minimizing the total operating cost within the current rolling optimization cycle. The total operating cost includes: energy consumption cost, unit start-up and shutdown switching cost, and spatial layout cost between units operating simultaneously.

[0174] The solver module 530 is used to solve the mixed integer nonlinear programming model so as to determine the unit start-stop combination and the load rate value allocated to each unit for each preset time step in the decision space composed of unit start-stop state variables and discretized load rate variables; and to adjust the load rate value of each unit under the condition of fixed unit start-stop combination to obtain the target load rate value of each unit.

[0175] The control module 540 is used to control the air source heat pump unit group from the current moment based on the unit start-stop combination and the target load rate value of each unit at a predetermined time step starting from the current moment, until the predetermined time step ends and the next rolling optimization cycle begins.

[0176] According to an embodiment of the present invention, the group control optimization device 500 further includes a constraint construction module. The constraint construction module is used to construct constraints for a mixed-integer nonlinear programming model. The constraints include: robust load balance constraints; the robust load balance constraints are configured to: ensure that the real-time total heat supply of the operating units is not less than the sum of the predicted heat load value at the corresponding time in the predicted heat load value sequence and a first safety margin based on confidence level; and ensure that the maximum total heating capacity of the operating units is not less than the sum of the predicted heat load value at the corresponding time in the predicted heat load value sequence and a second safety margin based on the statistical characteristics of prediction error, wherein the second safety margin is greater than the first safety margin.

[0177] According to an embodiment of the present invention, the group control optimization device 500 further includes: a first determining module, a second determining module, a third determining module, and a fourth determining module.

[0178] The first determination module determines the energy consumption cost based on the heating and electrical power of each unit and its start-up and shutdown status. The heating and electrical power is calculated based on a power model and operating parameters. The second determination module calculates the unit start-up and shutdown switching cost based on the change in the start-up and shutdown status of each unit at adjacent times. The third determination module calculates the spatial layout cost based on the reciprocal of the installation distance between any two simultaneously operating units. The fourth determination module sums the energy consumption cost, unit start-up and shutdown switching cost, and spatial layout cost to determine the total operating cost.

[0179] According to embodiments of this disclosure, any multiple modules among the prediction module 510, construction module 520, solving module 530, and control module 540 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functionality of one or more of these modules can be combined with at least some of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the prediction module 510, construction module 520, solving module 530, and control module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the prediction module 510, construction module 520, solution module 530, and control module 540 may be implemented at least partially as a computer program module that can perform corresponding functions when the computer program module is run.

[0180] It should be noted that the group control optimization device part for air source heat pump unit groups in the embodiments of the present invention corresponds to the group control optimization method part for air source heat pump unit groups in the embodiments of the present invention. For a detailed description of the group control optimization device part for air source heat pump unit groups, please refer to the group control optimization method part for air source heat pump unit groups, which will not be repeated here.

[0181] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0182] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A group control optimization method for air source heat pump unit groups, characterized in that, The group control optimization method includes: For the current rolling optimization cycle, starting from the current time and spaced N times at consecutive preset time steps, perform the following steps, where N is an integer greater than or equal to 3: The historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group are input into the heating load prediction model, and the predicted heating load value sequence within the current rolling optimization cycle is output. Based on the predicted heating load value sequence, the operating parameters of each unit in the air source heat pump unit group, and the installation distance between every two units, a mixed integer nonlinear programming model is constructed with the goal of minimizing the total operating cost within the current rolling optimization cycle. The total operating cost includes: energy consumption cost, unit start-up and shutdown switching cost, and spatial layout cost between simultaneously operating units. The mixed-integer nonlinear programming model is solved to determine the unit start-up / shutdown combinations and the assigned load factor values ​​for each unit within a decision space composed of unit start-up / shutdown state variables and discretized load factor variables, based on the preset time steps. Furthermore, with the unit start-up / shutdown combinations fixed, the load factor values ​​for each unit are adjusted to obtain the target load factor values ​​for each unit. Based on the unit start-up and shutdown combination and the target load rate value of each unit at a predetermined time step starting from the current time, the air source heat pump unit group is controlled from the current time until the predetermined time step ends and the next rolling optimization cycle begins.

2. The group control optimization method according to claim 1, characterized in that, The process of inputting the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group into the heating load prediction model, and outputting the predicted heating load value sequence within the current rolling optimization cycle, includes: The historical heating load value sequence and the historical environmental parameter value sequence are aligned with the time periods within the historical time periods, and then combined to obtain the first matrix; Align the estimated environmental parameter value sequence with the time within the current rolling optimization cycle and combine them to obtain a second matrix, in which the heating load value is empty; The first matrix and the second matrix are concatenated in chronological order to obtain the input matrix; The input matrix is ​​subjected to embedding mapping and position encoding to obtain a temporal embedding representation; Using a convolutional neural network, local features for characterizing fluctuation and periodic patterns are extracted from the temporal embedding representation; By utilizing a multi-head self-attention mechanism, dependencies are determined from the temporal embedding representation to obtain global features; The local features and the global features are fused to obtain fused features, and the features in the fused features that correspond to the current rolling optimization cycle are mapped to obtain the predicted heating load value sequence within the current rolling optimization cycle.

3. The group control optimization method according to claim 1 or 2, characterized in that, The group control optimization method further includes: constructing the constraints of the mixed integer nonlinear programming model; The constraints include: robust load balance constraints; The robust load balance constraint is configured as follows: The real-time total heat supply of the operating units shall not be less than the sum of the predicted heat load value at the corresponding time in the predicted heat load value sequence and the first safety margin based on confidence level. The maximum total heating capacity of the operating units shall not be less than the sum of the predicted heating load value at the corresponding time in the predicted heating load value sequence and the second safety margin based on the statistical characteristics of the prediction error, wherein the second safety margin is greater than the first safety margin.

4. The group control optimization method according to claim 3, characterized in that, The constraints also include minimum running time constraints and minimum downtime constraints; The minimum running time constraint is configured to limit the shortest duration of continuous operation of the unit after startup; The minimum downtime constraint is configured to limit the shortest duration of continuous downtime after the unit is shut down.

5. The group control optimization method according to claim 3, characterized in that, The constraints also include at least one of the following: operating status constraints, load rate constraints, and switching frequency constraints; The operating status constraints are configured to constrain the operating status of each unit based on the rated heating capacity and the predicted total heating load of each unit. The predicted total heating load is the sum of each predicted heating load value in the predicted heating load value sequence. The load rate constraint is configured to constrain the load rate of each unit based on the load rate range of each unit; The switching frequency constraint is configured to constrain the switching frequency of each unit's operating status based on a switching frequency threshold within the current rolling optimization cycle.

6. The group control optimization method according to claim 5, characterized in that, The load factor range includes an upper limit and a lower limit. For each generating unit, the upper limit of the load factor is determined according to the following steps: Based on the operating parameters and rated power parameters of the unit, a power model is constructed in the form of a quadratic function. Based on the relationship between the heating power and the load rate determined by the power model, the load rate value is determined when the difference between the power of the unit and the power set value meets the preset conditions. The load rate value is determined as the upper limit of the load rate.

7. The group control optimization method according to claim 1, characterized in that, The group control optimization method also includes: The energy cost is determined based on the heating and electrical power of each unit and the start-up and shutdown status of each unit. The heating and electrical power is calculated based on the power model and the operating parameters. The start-stop switching cost of the unit is calculated based on the change in start-stop status of each unit at adjacent times; The spatial layout cost is calculated based on the reciprocal of the installation distance between any two units operating simultaneously. The total operating cost is determined by summing the energy consumption cost, the unit start-up and shutdown switching cost, and the spatial layout cost.

8. The group control optimization method according to claim 1, characterized in that, The decision space includes multiple candidate solutions, which indicate the candidate unit start-up and shutdown combinations of each unit within each preset time step, as well as the candidate load rate values ​​pre-allocated to each unit corresponding to the candidate unit start-up and shutdown combinations. The determination of unit start-up / shutdown combinations and assigned load factor values ​​for each unit within a decision space comprised of unit start-up / shutdown state variables and discretized load factor variables at each preset time step includes: Using the discrete particle swarm optimization algorithm, the target candidate solution is iteratively determined from multiple candidate solutions in the decision space based on the constraints of the mixed integer nonlinear programming model; The candidate unit start-up and shutdown combinations within each preset time step indicated by the target candidate solution are determined as the unit start-up and shutdown combinations corresponding to the preset time step. The candidate load rate value corresponding to the candidate unit start-stop combination is determined as the load rate value.

9. The group control optimization method according to claim 8, characterized in that, The step of adjusting the load rate value of each unit to obtain the target load rate value of each unit while keeping the unit start-stop combination fixed includes: Based on the load rate values ​​of each unit, the load rate variable is treated as a continuous variable to determine the candidate value range for each load rate value; The load factor value of each unit is iteratively adjusted using the gradient descent algorithm. In each iteration, the gradient of the total operating cost with respect to the load factor variable is calculated, and the load factor value of each unit is updated along the descent direction that satisfies the constraint condition. Once the convergence condition is met, the iteration is stopped, and the updated load rate values ​​of each unit are used to determine the corresponding target load rate value.

10. A group control optimization device for a group of air source heat pump units, characterized in that, The group control optimization device includes: For the current rolling optimization cycle, starting from the current time and spaced N times at consecutive preset time steps, execute the following module, where N is an integer greater than or equal to 3: The prediction module is used to input the historical heating load value sequence, historical environmental parameter value sequence, and estimated environmental parameter value sequence within the current rolling optimization cycle of the air source heat pump unit group into the heating load prediction model, and output the predicted heating load value sequence within the current rolling optimization cycle. The construction module is used to construct a mixed integer nonlinear programming model based on the predicted heating load value sequence, the operating parameters of each unit in the air source heat pump unit group, and the installation distance between every two units, with the optimization objective of minimizing the total operating cost within the current rolling optimization cycle. The total operating cost includes: energy consumption cost, unit start-up and shutdown switching cost, and spatial layout cost between simultaneously operating units. The solution module is used to solve the mixed integer nonlinear programming model so as to determine the unit start-up and shutdown combinations and the load rate values ​​allocated to each unit in the decision space composed of unit start-up and shutdown state variables and discretized load rate variables; and to adjust the load rate values ​​of each unit while fixing the unit start-up and shutdown combinations to obtain the target load rate values ​​of each unit. The control module is used to control the air source heat pump unit group from the current moment based on the unit start-stop combination and the target load rate value of each unit within a predetermined time step starting from the current moment, until the predetermined time step ends and the next rolling optimization cycle begins.

Images