A pipe network balance simulation optimization method and device based on digital twinning and a medium

Abstract

The application discloses a pipe network balance simulation optimization method and device based on digital twinning, equipment and medium, relates to the technical field of computing modeling, and comprises the following steps: constructing a twinning object mapping table and solidifying pipe network twinning control parameters, generating real-time twinning control states by performing discrete coding and consistency checking, performing control state evolution update on the real-time twinning control states by a physical information neural network, and screening to obtain a decision twinning working condition; simulating balance evolution based on the decision twinning working condition, and identifying a control influence chain set causing imbalance by a key path reverse tracing topology algorithm; performing control evolution simulation based on the control influence chain set and the decision twinning working condition, and performing comprehensive evaluation to generate a strategy evaluation set; performing adjustment operations according to the strategy evaluation set, and writing back to the real-time twinning control state to generate a pipe network balance operation confirmation. The application realizes intelligent closed-loop optimization of pipe network balance.

Description

Technical Field

[0001] This invention relates to the field of computational modeling technology, and in particular to a method, equipment, and medium for pipeline network balance simulation and optimization based on digital twins. Background Technology

[0002] In the operation and management of pipeline networks (such as water supply and heating networks), balance simulation optimization is a crucial step in ensuring hydraulic stability and energy efficiency. Conventional methods are typically based on physical information models, such as numerical solutions to the Saint-Venant equations, combined with historical operation and maintenance data to calibrate parameters and perform offline simulations of actuators such as valves and pumps. These methods discretize the pipeline network topology and apply optimization algorithms (such as gradient descent or genetic algorithms) to adjust control parameters, aiming to maintain pressure and flow equilibrium. With the development of digital twin technology, its application in industry is becoming increasingly widespread, providing the possibility of virtual mapping for pipeline network management.

[0003] However, conventional methods have certain limitations in practice. On the one hand, simulation accuracy is limited by fixed parameters and simplifying assumptions, making it difficult to efficiently capture nonlinear effects and uncertainties in the dynamic behavior of pipeline networks (such as measurement noise and transient flow field changes), leading to deviations between predicted states and actual operation, and affecting the reliability of evolution updates. On the other hand, the optimization process is insufficient in the coordinated handling of multi-objective constraints (such as balance restoration, energy consumption control, and control action smoothness), and the algorithm is prone to getting trapped in local optima or slow response, weakening the practicality and stability of decision-making conditions. Digital twin methods can alleviate these shortcomings through intelligent computing, but traditional methods often lack systematic adaptation when integrating advanced algorithms. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a pipeline network balancing simulation optimization method based on digital twins to solve the problems of inaccurate real-time multi-objective decision-making and simulation adaptability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a pipeline network balancing simulation optimization method based on digital twins, comprising: constructing a twin object mapping table and fixing pipeline network twin control parameters; generating a real-time twin control state by performing discrete coding and consistency verification; performing control state evolution updates on the real-time twin control state through a physical information neural network and filtering to obtain decision twin operating conditions; simulating the balancing evolution based on the decision twin operating conditions and identifying the set of control influence chains leading to imbalance through a critical path back-tracing topology algorithm; performing control evolution simulation based on the set of control influence chains and the decision twin operating conditions, and performing a comprehensive evaluation to generate a strategy evaluation set; performing adjustment operations according to the strategy evaluation set and writing back to the real-time twin control state to generate a pipeline network balancing operation confirmation.

[0007] As a preferred embodiment of the pipeline network balancing simulation and optimization method based on digital twins described in this invention, the steps of constructing the twin object mapping table and fixing the pipeline network twin control parameters are as follows: Based on the pipeline network topology data, a twin object mapping table is constructed according to the control object mapping relationship of node-pipe segment-execution component; Based on the twin object mapping table, roles are fixed according to the separation principle, and pipeline twin control parameters are generated.

[0008] As a preferred embodiment of the pipeline network balancing simulation and optimization method based on digital twins described in this invention, the step of generating a real-time twin control state by performing discrete coding and consistency verification is as follows: Discretely encode various parameters in the pipeline twin control parameters to generate the initial twin control state; The initial twin control state is checked for interface consistency and timestamp consistency to filter out the real-time twin control state.

[0009] As a preferred embodiment of the pipeline network balancing simulation optimization method based on digital twins described in this invention, the steps of performing control state evolution updates on the real-time twin control state through a physical information neural network and selecting decision twin operating conditions are as follows: The trained physical information neural network is obtained by iteratively updating the physical information neural network by minimizing the loss function. The real-time twin control state is updated by performing control state evolution through a trained physical information neural network to generate real-time twin operating conditions. The control stability and constraint conditions of the real-time twin operating conditions are determined, and the decision twin operating conditions are obtained by screening through a multi-objective optimization algorithm.

[0010] As a preferred embodiment of the pipeline network balance simulation and optimization method based on digital twins described in this invention, the steps of simulating the balance evolution based on decision twin operating conditions and identifying the set of control influence chains leading to imbalance through a critical path back-tracing topology algorithm are as follows: Based on decision twin operating conditions, the equilibrium evolution of the pipeline network during operation is calculated by recursively simulating the control state, thus forming a twin equilibrium evolution set; By comparing the set of twin equilibrium evolutions with the hydraulic equilibrium constraints, the set of twin imbalance control states is identified. By treating nodes, pipe segments, and execution components as nodes and using node-pipe segment connections as edges, an influence propagation matrix is ​​constructed. The Dijkstra algorithm is used to reverse the propagation path of influence, and the propagation path of control state is traced back to locate the set of control influence chains that lead to the imbalance.

[0011] As a preferred embodiment of the pipeline network balancing simulation optimization method based on digital twins described in this invention, the steps of performing control evolution simulation and generating a strategy evaluation set based on the control influence chain set and decision twin operating conditions are as follows: Construct a control action adjustment space at the corresponding control node of the control influence chain set to generate a candidate control strategy set; The candidate control strategy set is loaded into the decision twin operating condition, and the control evolution simulation is performed using a multi-objective constrained optimization algorithm to obtain the strategy evolution set; Based on the degree of balance recovery, the amount of energy consumption change, and the scale of control action changes, a comprehensive evaluation of the strategy evolution set is conducted to generate a strategy evaluation set. The comprehensive evaluation includes normalizing the balance recovery degree, energy consumption change, and control action execution to obtain a base score, and mapping the constraint violation degree as a penalty factor to be added to the base score for correction.

[0012] As a preferred embodiment of the pipeline network balancing simulation optimization method based on digital twins described in this invention, the step of performing adjustment operations according to the strategy evaluation set includes the following steps: The control state change relationships defined in the strategy evaluation set are expanded into a sequence of control actions and aggregated to form a balanced control execution sequence. According to the temporal dependencies in the balance control execution sequence, the valves and pumps are driven to perform regulation operations in sequence, generating the pipeline control execution trajectory.

[0013] As a preferred embodiment of the pipeline network balancing simulation and optimization method based on digital twins described in this invention, the step of writing back to the real-time twin control state to generate pipeline network balancing operation confirmation includes the following steps: Extract the valve opening status and pump operating parameter change values ​​corresponding to the pipeline control execution trajectory, and write them back to the real-time twin control status through the status synchronization buffer queue to update and form a closed-loop twin control status; The closed-loop twin control status is judged based on the criteria for achieving pipeline balance, and a pipeline balance operation confirmation is generated.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the pipeline network balancing simulation optimization method based on digital twins as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the pipeline network balancing simulation optimization method based on digital twins as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by using a physical information neural network to perform control state evolution updates and state deduction on the real-time twin control state, accurate simulation of the dynamic behavior of the pipeline network is achieved; by using a critical path reverse tracing topology algorithm to accurately identify the imbalance propagation path, the insight and decision reliability of the digital twin are enhanced, and intelligent closed-loop optimization of pipeline network balance is realized. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0018] Fig. 1 This is a flowchart of a pipeline network balancing simulation optimization method based on digital twins.

[0019] Fig. 2 A flowchart for constructing real-time twin control states.

[0020] Fig. 3 A flowchart for generating decision twin operating conditions.

[0021] Fig. 4 A flowchart for identifying control influence chains and generating a set of policy evaluations. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figs. 1-4 As one embodiment of the present invention, this embodiment provides a pipeline network balancing simulation optimization method based on digital twins, comprising the following steps: S1. Construct a twin object mapping table and solidify the twin control parameters of the pipeline network. Generate real-time twin control status by performing discrete coding and consistency verification.

[0026] Based on the pipeline network topology data, a twin object mapping table is constructed according to the control object mapping relationship of node-pipe segment-execution component.

[0027] Furthermore, the system reads the as-built drawings and historical operation and maintenance records, extracts the node identifier, pipe segment identifier, node-pipe segment connection relationship, and flow direction marker fields, and generates pipeline network topology data. Based on the node-pipe segment connection relationship, a directed connection matrix is ​​generated. Using the directed connection matrix as a structural constraint, the system traverses the pipe segment identifier associated with each node and binds the execution components (valves and pumps) connected to the pipe segment to the corresponding node-pipe segment combination position. The node number, pipe segment number, and execution component number are sequentially encoded. After encoding, the one-to-one correspondence between nodes, pipe segments, and execution components is written into an index table and a twin object mapping table is formed.

[0028] It should be noted that the historical operation and maintenance records include historical flow sampling values, historical pressure sampling values, historical measurement resolution, historical measurement noise, historical energy conservation residual data sequences, historical steady-state residual data sequences, and equipment identification fields. Heterogeneous data are uniformly associated by using the component number as the primary key.

[0029] Based on the twin object mapping table, roles are fixed according to the separation principle, and pipeline twin control parameters are generated.

[0030] Furthermore, based on the twin object mapping table, the node-pipe segment correspondence is used as the state object and the execution component is used as the adjustment object. The parameter fields in the twin object mapping table are assigned roles. The fields representing the pipe segment structure and flow direction attributes are solidified into a set of state parameters, and the fields representing the valve opening value range and the pump operation parameter value range are solidified into a set of control parameters (e.g., the valve opening is limited to 0 to 1, and the pump operation parameter is limited to 800 to 2900 revolutions per minute). The set of state parameters and the set of control parameters are organized according to a unified field order to generate the pipeline twin control parameters for subsequent discrete coding and control state evolution calculation.

[0031] It should be noted that the separation principle refers to dividing the parameter fields into an immutable set of state parameters and a controllable set of control parameters based on the differences between the pipeline network topology attributes (including node identifiers, pipe segment identifiers, and node-pipe segment connection relationships) and the adjustment attributes of the actuators (referring to the action boundaries of actuators such as valves and pumps) when constructing the twin object mapping table, so as to achieve the clear and solidified object role.

[0032] The various parameters in the pipeline twin control parameters are discretely encoded to generate the initial twin control state.

[0033] Furthermore, based on the pipeline twin control parameters, the set of state parameters is discretized using interval equidistant quantization rules (e.g., flow direction markings are kept binary encoded, and pipe segment resistance coefficients are discretized into finite levels with a step size of 0.01). At the same time, the set of control parameters is quantized into intervals and mapped to levels (e.g., the valve opening value range of 0 to 1 is divided into 21 discrete values ​​with a step size of 0.05, and the pump operating parameter value range of 800 to 2900 revolutions per minute is divided into discrete levels with a step size of 100). After all parameters are discretized, the discretized pipeline twin control parameters are mapped into control state vectors according to the unified index order of node number, pipe segment number, and execution component number, and then spliced ​​to generate the initial twin control state.

[0034] It should be noted that the interval equidistant quantization rule uses the minimum adjustable step size (e.g., 0.01) of the valve and pump actuators and the historical measurement resolution as the setting basis, and determines the quantization step size within the value range according to "an integer multiple (e.g., 2 times) of the minimum adjustable step size and the measurement resolution" (the 0.05 step size in the example is a further set step size after combining the equipment control level; if the minimum adjustable step size is 0.01, then 0.05 is a 5-fold relationship, which conforms to the integer multiple rule); the discrete level refers to the set of finite range values ​​obtained by dividing the continuous value parameters in the pipeline twin control parameters according to the interval equidistant quantization rule. For example, the valve opening value range 0 to 1 is divided into 21 discrete values ​​with a step size of 0.05 to form the discrete level of the valve opening state.

[0035] The initial twin control state is checked for interface consistency and timestamp consistency to filter out the real-time twin control state.

[0036] Furthermore, the execution component number and interface identifier fields in the initial twin control state are compared one by one with the device identifier fields in the historical operation and maintenance records. After establishing a mapping relationship between the interface identifier fields and device identifier fields based on the execution component number, consistency verification is performed. When the interface identifier field and device identifier field are completely consistent, the corresponding control state record is retained. When the interface identifier field is missing or inconsistent, the corresponding control state record is removed. After completing the interface consistency verification, the timestamp field of the retained control state records is read and sorted in ascending order of time. When the difference between adjacent timestamps is less than the synchronization window (e.g., 2 seconds), it is determined to be a valid synchronization state. When the difference between timestamps exceeds the synchronization window, the lagging record is deleted. The control state record that simultaneously satisfies interface consistency and timestamp consistency is written into the real-time twin control state.

[0037] S2. The control state evolution update is performed on the real-time twin control state through the physical information neural network, and the decision twin operating conditions are obtained by screening.

[0038] The trained physical information neural network is obtained by iteratively updating the physical information neural network by minimizing the loss function.

[0039] The physical information neural network is trained in two stages. In the first stage, transient flow field simulation data is pre-generated based on pipeline network topology data. The mean square error of state prediction and the weighted loss value are obtained through a loss function (including the mean square error of state prediction and the residual term of the one-dimensional Saint-Venant equation). The Adam optimizer is used to perform backpropagation to update the network parameters based on the weighted loss value with a learning rate (e.g., 0.001) and a batch size (e.g., 32). Iterative training generates physical information neural network parameters that initially fit the hydraulic law of the pipeline network. In the second stage, historical flow and pressure sampling values ​​from historical operation and maintenance records are used for fine-tuning. A node flow conservation constraint term is added to the loss function. The Adam optimizer is used to verify that the loss changes for multiple consecutive rounds (e.g., 5 rounds) when the magnitude is less than the loss threshold, and the trained physical information neural network is obtained.

[0040] It should be noted that the loss threshold is set based on the standard deviation multiple (e.g., 1.5 times) of the relative rate of change of the validation loss in the most recent consecutive iterations (e.g., the most recent 10 iterations) during training. An exemplary value range is typically between 0.05% and 0.5%.

[0041] The real-time twin control state is updated by performing control state evolution through a trained physical information neural network, thereby generating real-time twin operating conditions.

[0042] Furthermore, the real-time twin control state is input into the trained physical information neural network, which includes a graph attention propagation layer and LSTM filtering. Forward inference is performed using a basic time step (e.g., 0.01 seconds), and the pressure gradient magnitude between nodes is calculated in real time to refine the local time step. When the energy conservation residual value exceeds the conservation residual threshold after a single-step inference, the smaller value between the weighted energy conservation residual value (i.e., the energy conservation residual value multiplied by a fixed coefficient, e.g., 0.25) and the upper limit of the correction coefficient (e.g., 0.5) is taken as the correction state vector. The correction state vector is input into the graph attention propagation layer, and attention weights are obtained through softmax normalization and weighted aggregation to generate the propagation state vector. The propagation state vector is superimposed with the corresponding historical measurement noise from the historical operation and maintenance records (e.g., the amplitude is 1.8 times the measurement resolution), filtered by LSTM, and the corrected Darcy friction factor is output. The parameters of the physical information neural network are updated based on the corrected Darcy friction factor as the input for the next loop. After completing the inference for a specified number of time steps (e.g., 100), the corresponding converged state vector is extracted to generate the real-time twin operating condition.

[0043] It should be noted that the energy conservation residual refers to the inconsistency between the pressure difference between the two ends of the pipe section and the friction loss and local loss. The energy conservation residual is obtained by taking the pressure value, flow value and hydraulic parameters of the pipe section in each recursive step through real-time twin control. The energy conservation residual threshold is set according to the fixed quantile value (e.g., 95th percentile) of the historical energy conservation residual data sequence under steady-state conditions in the historical operation and maintenance records. The exemplary value range is usually 0.5% to 10%.

[0044] The control stability and constraint conditions of the real-time twin operating conditions are determined, and the decision twin operating conditions are obtained by screening through a multi-objective optimization algorithm.

[0045] Furthermore, convergent state vectors are extracted from real-time twin operating conditions, and the difference norm of adjacent convergent state vectors is obtained as a control stability index. When the control stability index exceeds the stability threshold or the energy conservation residual exceeds the conservation residual threshold, the corresponding real-time twin operating condition is deleted. Real-time twin operating conditions that simultaneously satisfy the stability threshold and the conservation residual threshold are retained to form a feasible operating condition set. A multi-objective optimization algorithm is used on the feasible operating condition set with equilibrium recovery deviation, energy consumption change, and control action change scale as three objective vectors. A non-dominated sorting genetic algorithm is used for iterative updates and outputs the Pareto front (e.g., the population size is 64 to 128, and the number of iterations is 30 to 80). The minimum equilibrium recovery deviation and the energy consumption change not exceeding the energy consumption threshold are used as selection conditions, and the control action change scale is used as a secondary sorting index to screen and obtain decision twin operating conditions.

[0046] It should be noted that the multi-objective optimization algorithm is a screening tool used to balance the three objectives of equilibrium recovery, energy consumption, and control action, and output the optimal Pareto front solution from feasible operating conditions; the stability threshold is set based on the fixed percentile (e.g., the 95th percentile) of the difference norm statistical distribution of the control state vector in historical operation and maintenance records, and the exemplary value range is usually 0.03 to 0.10; the energy consumption threshold is set based on the statistical distribution of historical energy consumption changes, and the exemplary value range is usually 1% to 5%.

[0047] The expression for calculating the difference norm of adjacent convergent state vectors is: ; in, For time indexing; The L2 norm is the difference between the convergent state vectors at adjacent time points; For a moment The convergent state vector; For a moment The convergent state vector.

[0048] S3. Based on the decision twin operating conditions, simulate the equilibrium evolution situation, and identify the set of control influence chains that lead to imbalance through the critical path back-tracing topology algorithm.

[0049] Based on decision twin operating conditions, the equilibrium evolution of the pipeline network during operation is calculated by recursively simulating the control state, thus forming a twin equilibrium evolution set.

[0050] Furthermore, based on the decision twin operating conditions, a trained physical information neural network is used as the state transition function, and forward recursion is performed at a basic time step (e.g., 0.01 seconds) (e.g., the number of recursion steps is 100 to 500). In each recursion step, the pressure gradient magnitude between nodes is calculated, and the local time step is determined according to the pressure gradient threshold set. Sub-step recursion is performed on the pipe segments hit by the local time step set. After the sub-step recursion is completed, a weighted least squares objective is constructed with the mass conservation residual and the energy conservation residual, and the projected state vector is solved under the constraints of the valve opening value range and the pump operating parameter value range (e.g., the number of iterations is 8, and the damping coefficient is 0.6). The projected state vector is then aggregated to form a twin equilibrium evolution set.

[0051] It should be noted that the pressure gradient threshold set is set based on the fixed quantile of the statistical distribution of historical pressure gradient amplitudes. For example, the pressure gradient threshold set takes 0.005 MPa per meter as the first threshold and 0.020 MPa per meter as the second threshold. When the pressure gradient amplitude is less than the first threshold, it is classified as a low gradient, with a local time step of 0.01 seconds and a sub-step recursion count of 1. When the pressure gradient amplitude is greater than or equal to the first threshold but less than the second threshold, it is classified as a medium gradient, with a local time step of 0.005 seconds and a sub-step recursion count of 2. When the pressure gradient amplitude is greater than or equal to the second threshold, it is classified as a high gradient, with a local time step of 0.002 seconds and a sub-step recursion count of 5. The mass conservation residual refers to the unbalanced amount after the inflow and outflow at the node are superimposed with the node's water consumption. It is obtained by summarizing the flow values ​​of the decision twin working condition and the node-pipe connection relationship in each recursion step. The projected state vector is the corrected state after satisfying the constraints, which is used to replace the predicted value of the forward recursion to ensure hydraulic consistency.

[0052] By comparing the set of twin equilibrium evolutions with the hydraulic equilibrium constraints, the set of twin imbalance control states is identified.

[0053] Furthermore, based on the projected state vectors in the twin equilibrium evolution set, the mass conservation residuals and energy conservation residuals are concatenated in a unified index order to form a constraint residual vector; using the historical steady-state residual data sequence from historical operation and maintenance records, the constraint residual vectors are transposed and the Mahalanobis distance of the constraint residual vectors is obtained; projected state vectors with Mahalanobis distances greater than the imbalance judgment threshold are identified and marked as imbalance marker projected state vectors; continuity discrimination is performed on the imbalance marker projected state vectors according to the sliding window length and the number of consecutive threshold exceedances (e.g., the sliding window length is 10 and the number of consecutive threshold exceedances is 3), and the twin imbalance control state set is output by collecting the imbalance marker projected state vectors through continuity discrimination.

[0054] It should be noted that the imbalance determination threshold is set based on a fixed percentage (e.g., 95%) of the historical Mahalanobis distance statistical distribution, with an exemplary range typically from 10 to 30.

[0055] The expression for calculating Mahalanobis distance is: ; in, To constrain the Mahalanobis distance of the residual vector; To constrain the residual vector; This is the transpose of the constraint residual vector.

[0056] By treating nodes, pipe segments, and execution components as nodes and using node-pipe segment connections as edges, an influence propagation matrix is ​​constructed.

[0057] Furthermore, the node identifier, pipe segment identifier, and execution component number are concatenated into a graph node set to construct a heterogeneous directed graph composed of three types of nodes: nodes, pipe segments, and execution components. A set of structural edges is generated based on the node-pipe segment connection relationship. The trained physical information neural network is used to perform partial derivative calculations on the control parameter set at the decision twin operating condition to obtain the Jacobian matrix of node pressure value and pipe segment flow value with respect to valve opening value range and pump operating parameter value range. The influence intensity of the Jacobian matrix is ​​mapped to the edge weights of execution "component number → pipe segment number", "pipe segment number → node number", and "node number → pipe segment number". Multiple edge weights are retained row by row and normalized and written into a sparse matrix (e.g., retaining the first 3) to generate the influence propagation matrix.

[0058] The Dijkstra algorithm is used to reverse the propagation path of influence, and the propagation path of control state is traced back to locate the set of control influence chains that lead to the imbalance.

[0059] Furthermore, Dijkstra's algorithm is executed to backtrack backward based on the direction of the edges in the influence propagation matrix. Specifically, starting from the imbalance node corresponding to the set of twin imbalance control states, candidate predecessor nodes are expanded layer by layer along the incoming edge direction of the influence propagation matrix. At each expansion, the edge weight is obtained as the cumulative cost based on the decision twin operating conditions. The edge weight is obtained by normalizing the coupling term of the pressure gradient magnitude between nodes, the valve opening state change magnitude, and the pump operating parameter change magnitude. The predecessor chain with the minimum cumulative cost is relaxed and updated until it backtracks to the corresponding node of the execution component, forming a path sequence. The path sequence is truncated and aggregated according to the cumulative cost from small to large to output the control influence chain set, which is used to accurately identify the imbalance propagation path to locate the root cause of the imbalance, and to provide a basis for constructing a set of candidate control strategies at the corresponding control nodes.

[0060] It should be noted that the valve opening status change amplitude refers to the absolute value of the difference between the valve opening values ​​of the valve corresponding to the same actuator number at adjacent control times; the pump operating parameter change amplitude refers to the absolute value of the difference between the pump operating parameter values ​​of the pump corresponding to the same actuator number at adjacent control times.

[0061] S4. Based on the control influence chain set and decision twin operating conditions, perform control evolution simulation, conduct comprehensive evaluation, and generate a strategy evaluation set.

[0062] At the corresponding control node of the control influence chain set, a control action adjustment space is constructed to generate a candidate control strategy set.

[0063] Furthermore, using the control nodes corresponding to the execution components in the control influence chain set as indices, control action adjustment spaces of "hold", "adjust one level", and "adjust one level" are constructed at each execution component (for example, the valve opening state is moved up or down by one level according to discrete values ​​from 0 to 1, and the pump operating parameters are moved up or down by one level according to discrete levels from 800 to 2900 revolutions per minute). The control action adjustment space is searched for combinations of execution bundles based on the cumulative cost of the influence propagation matrix, and the combinations with the minimum cost are truncated (for example, the bundle width is 16 and the combination depth is 3) to generate a candidate control strategy set. Each strategy in the candidate control strategy set contains the execution component number sequence and the corresponding control action adjustment space.

[0064] The candidate control strategy set is loaded into the decision twin operating condition, and the control evolution simulation is performed using a multi-objective constrained optimization algorithm to obtain the strategy evolution set.

[0065] Furthermore, the candidate control strategy set is loaded into the decision twin operating condition. Under the constraints of the valve opening value range and the pump operating parameter value range, the trained physical information neural network is used as the state transition function to perform forward recursive deduction for each candidate control strategy and output the twin equilibrium evolution set. The equilibrium recovery degree, energy consumption change and control action change scale are calculated through the twin equilibrium evolution set and concatenated to form the strategy objective vector. A multi-objective constraint optimization algorithm is used to construct the penalty term with the constraint residual vector Mahalanobis distance as the constraint violation degree. A constraint non-dominated sorting genetic algorithm is used to perform iterative updates on the candidate control strategy set (e.g., the number of iterations is 30 to 80), retaining feasible solutions and non-dominated solutions, and generating a strategy evolution set containing the candidate control strategy number and strategy objective vector, constraint violation degree, and twin equilibrium evolution set.

[0066] It should be noted that the degree of balance recovery refers to the decrease ratio of the norm of the constraint residual vector in the last frame of the twin balance evolution set relative to the norm of the constraint residual vector in the first frame; the change in energy consumption refers to the change ratio of the pump power integral value in the twin balance evolution set relative to the pump power integral value of the decision twin operating condition baseline, which is obtained by mapping the pump operating parameter values ​​to power values ​​and integrating them at time steps, then normalizing the difference with the baseline; the scale of control action change refers to the sum of the absolute values ​​of the differences between the candidate control strategy and the discrete levels of the decision twin operating condition in the sequence of execution component numbers; the multi-objective constraint optimization algorithm refers to the optimization method that iteratively evolves the candidate control strategy by introducing a constraint violation penalty term to ensure hydraulic balance constraints, focusing on strategy simulation evolution.

[0067] Based on the degree of balance recovery, the amount of energy consumption change, and the scale of control action changes, a comprehensive evaluation of the strategy evolution set is conducted to generate a strategy evaluation set.

[0068] Furthermore, minimum-maximum normalization is applied to the balance recovery degree, energy consumption change, and control action change scale. The weight coefficients of the balance recovery degree, energy consumption change, and control action change scale are solved by minimizing the pairwise sorting loss (e.g., the weight coefficients range from 0 to 1 and the weight sum is 1). The normalized balance recovery degree, energy consumption change, and control action change scale are weighted and aggregated according to the weight coefficients to obtain the basic score of the comprehensive evaluation. The constraint violation degree is mapped to a penalty factor and superimposed on the basic score of the comprehensive evaluation to obtain the comprehensive evaluation score (e.g., the penalty factor ranges from 0 to 0.3). Multiple candidate control strategy numbers (e.g., the top 10) are extracted according to the comprehensive evaluation scores and aggregated to output the strategy evaluation set.

[0069] S5. Execute adjustment operations based on the strategy evaluation set and write back to the real-time twin control status to generate a pipeline network balance operation confirmation.

[0070] The control state change relationships defined in the strategy evaluation set are expanded into a sequence of control actions, which are then aggregated to form a balanced control execution sequence.

[0071] Furthermore, the target candidate control strategy number is selected by sorting the comprehensive evaluation scores within the strategy evaluation set, and the discrete levels of valve opening status and pump operating parameters are located. The corresponding discrete levels within the decision twin operating conditions are used as references to generate control state change relationships. Based on the edge direction of the influence propagation matrix, the sequence of execution component numbers is topologically sorted to determine the sequence constraints of control actions, and the control state change relationships are mapped to a sequence of control actions with timestamp fields (e.g., the time interval between adjacent control actions is 0.2 seconds), and a balanced control execution sequence is generated.

[0072] According to the temporal dependencies in the balance control execution sequence, the valves and pumps are driven to perform regulation operations in sequence, generating the pipeline control execution trajectory.

[0073] Furthermore, based on the balanced control execution sequence, the valve and pump interface identifiers are located by the execution component number. The discrete levels of valve opening status and pump operating parameters are mapped to valve opening setpoints and pump operating parameter setpoints, and then executed sequentially. A fixed execution interval (e.g., 0.2 seconds) is maintained between adjacent control actions, and confirmation messages (including action acceptance flag, current valve opening feedback value, and pump operating parameter feedback value) are collected. The execution component number, setpoint, and issuance timestamp field and confirmation timestamp field are sequentially aggregated to generate the pipeline control execution trajectory.

[0074] Extract the corresponding valve opening status and pump operating parameter change values ​​from the pipeline control execution trajectory, and write them back to the real-time twin control status through the status synchronization buffer queue to update and form a closed-loop twin control status.

[0075] Furthermore, based on the pipeline control execution trajectory, the current valve opening status change value, pump operating parameter change value, and confirmation message are read according to the execution component number and confirmation timestamp field, and written into the status synchronization buffer queue. When the synchronization window is closed, the latest record in the synchronization window is retrieved according to the execution component number and written back to the valve opening status field and pump operating parameter field of the real-time twin control status to generate a closed-loop twin control status.

[0076] The closed-loop twin control status is judged based on the criteria for achieving pipeline balance, and a pipeline balance operation confirmation is generated.

[0077] Furthermore, the pressure deviation index is formed by reading the difference between the node pressure value and the minimum pressure from the closed-loop twin control state, and the current mass conservation residual and energy conservation residual are obtained. The mass conservation residual, energy conservation residual and pressure deviation index are synchronously input into the hydraulic balance achievement judgment conditions and a step-by-step comparison is performed to determine whether the closed-loop twin control state has achieved the pipeline balance achievement judgment conditions (for example, the node mass conservation residual limit is 0.5% to 2%, the pipeline section energy conservation residual limit is 1% to 5%, and the node pressure lower limit is 0.15MPa to 0.30MPa). When the pipeline balance achievement judgment conditions are met, a pipeline balance operation confirmation is generated.

[0078] This embodiment also provides a computer device applicable to the pipeline network balancing simulation optimization method based on digital twins, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the pipeline network balancing simulation optimization method based on digital twins as proposed in the above embodiment.

[0079] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0080] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the pipeline balancing simulation optimization method based on digital twins as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0081] In summary, this invention achieves accurate simulation of the dynamic behavior of the pipeline network by using a physical information neural network to perform control state evolution updates and state deductions on the real-time twin control state; and by accurately identifying imbalance propagation paths through a critical path back-tracing topology algorithm, it enhances the insight and decision reliability of the digital twin, and realizes intelligent closed-loop optimization of pipeline network balance.

[0082] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A pipeline network balancing simulation and optimization method based on digital twins, characterized in that, include: Construct a twin object mapping table and solidify the twin control parameters of the pipeline network. Generate real-time twin control status by performing discrete coding and consistency verification. The control state evolution update is performed on the real-time twin control state through a physical information neural network, and the decision twin operating conditions are obtained by filtering. Based on decision twin operating conditions, the equilibrium evolution is simulated, and the set of control influence chains leading to imbalance is identified through critical path back-tracing topology algorithm. Based on the control influence chain set and decision twin operating conditions, control evolution simulation is performed, and a comprehensive evaluation is conducted to generate a strategy evaluation set. Adjustment operations are performed based on the strategy evaluation set and written back to the real-time twin control state to generate a pipeline network balance operation confirmation.

2. The pipeline network balancing simulation and optimization method based on digital twins as described in claim 1, characterized in that, The steps for constructing the twin object mapping table and fixing the twin control parameters of the pipeline network are as follows: Based on the pipeline network topology data, a twin object mapping table is constructed according to the control object mapping relationship of node-pipe segment-execution component; Based on the twin object mapping table, roles are fixed according to the separation principle, and pipeline twin control parameters are generated.

3. The pipeline network balancing simulation and optimization method based on digital twins as described in claim 2, characterized in that, The process of generating a real-time twin control state by performing discrete coding and consistency verification is as follows: Discretely encode various parameters in the pipeline twin control parameters to generate the initial twin control state; The initial twin control state is checked for interface consistency and timestamp consistency to filter out the real-time twin control state.

4. The pipeline network balancing simulation and optimization method based on digital twins as described in claim 1, characterized in that, The steps for performing control state evolution updates on the real-time twin control state through a physical information neural network and filtering to obtain decision twin operating conditions are as follows: The trained physical information neural network is obtained by iteratively updating the physical information neural network by minimizing the loss function. The real-time twin control state is updated by performing control state evolution through a trained physical information neural network to generate real-time twin operating conditions. The control stability and constraint conditions of the real-time twin operating conditions are determined, and the decision twin operating conditions are obtained by screening through a multi-objective optimization algorithm.

5. The pipeline network balancing simulation and optimization method based on digital twins as described in claim 1, characterized in that, The steps for simulating the equilibrium evolution based on decision twin operating conditions and identifying the set of control influence chains leading to imbalance through a critical path back-tracing topology algorithm are as follows: Based on decision twin operating conditions, the equilibrium evolution of the pipeline network during operation is calculated by recursively simulating the control state, thus forming a twin equilibrium evolution set; By comparing the set of twin equilibrium evolutions with the hydraulic equilibrium constraints, the set of twin imbalance control states is identified. By treating nodes, pipe segments, and execution components as nodes, and using the control influence relationships between nodes, pipe segments, and execution components as edges, an influence propagation matrix is ​​constructed. The Dijkstra algorithm is used to reverse the propagation path of influence, and the propagation path of control state is traced back to locate the set of control influence chains that lead to the imbalance.

6. The pipeline network balancing simulation and optimization method based on digital twins as described in claim 1, characterized in that, The steps for performing control evolution simulation and generating a strategy evaluation set based on the control influence chain set and decision twin operating conditions are as follows: Construct a control action adjustment space at the corresponding control node of the control influence chain set to generate a candidate control strategy set; The candidate control strategy set is loaded into the decision twin operating condition, and the control evolution simulation is performed using a multi-objective constrained optimization algorithm to obtain the strategy evolution set; Based on the degree of balance recovery, the amount of energy consumption change, and the scale of control action changes, a comprehensive evaluation of the strategy evolution set is conducted to generate a strategy evaluation set. The comprehensive evaluation includes normalizing the balance recovery degree, energy consumption change, and control action execution to obtain a base score, and mapping the constraint violation degree as a penalty factor to be added to the base score for correction.

7. The pipeline network balancing simulation and optimization method based on digital twins as described in claim 1, characterized in that, The steps for performing the adjustment operation based on the policy evaluation set are as follows: The control state change relationships defined in the strategy evaluation set are expanded into a sequence of control actions and aggregated to form a balanced control execution sequence. According to the temporal dependencies in the balance control execution sequence, the valves and pumps are driven to perform regulation operations in sequence, generating the pipeline control execution trajectory.

8. The pipeline network balancing simulation and optimization method based on digital twins as described in claim 7, characterized in that, The steps for writing back to the real-time twin control state and generating pipeline network balanced operation confirmation are as follows: Extract the corresponding valve opening status and pump operating parameter change values ​​from the pipeline control execution trajectory, and write them back to the real-time twin control status through the status synchronization buffer queue to update and form a closed-loop twin control status; The closed-loop twin control status is judged based on the criteria for achieving pipeline balance, and a pipeline balance operation confirmation is generated.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the pipeline network balancing simulation optimization method based on digital twins as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the pipeline network balancing simulation optimization method based on digital twins as described in any one of claims 1 to 8.

Images