An urban rail energy consumption scheduling method based on artificial intelligence multi-objective optimization

Abstract

The application discloses a kind of urban rail energy consumption scheduling methods based on artificial intelligence multi-objective optimization, comprising: by comprehensive real-time data processing, multi-objective collaborative optimization and intelligent algorithm iteration solution, effectively solve the scheduling decision limitation problem brought by single optimization target in prior art, realize the optimal control of whole operation energy consumption of rail transit system;Effectively improve train operation energy efficiency, regenerative energy utilization rate and operation punctuality rate, while reducing the peak load of traction power supply system, significantly improve the overall energy utilization efficiency and operation performance of rail transit, realize the dual goal of system energy saving and efficient operation.

Description

Technical Field

[0001] This invention relates to the technical field of urban rail transit operation scheduling, and in particular to an urban rail energy consumption scheduling method based on artificial intelligence multi-objective optimization. Background Technology

[0002] Urban rail transit, with its high efficiency, low pollution, and large capacity, has become an important infrastructure for modern metropolises to alleviate traffic pressure and promote socio-economic development. However, with the rapid expansion of urban rail transit networks and the continuous increase in passenger volume, the energy consumption of train operation has gradually become prominent, becoming one of the key factors restricting the sustainable development of rail transit. At present, urban rail transit energy consumption optimization and scheduling technologies mainly focus on optimizing train departure intervals, regenerative braking energy recovery and utilization, and dynamic passenger flow response.

[0003] CN113743828A discloses a method for urban rail transit operation scheduling, which focuses on modeling the DC traction power supply system and integrating regenerative braking energy utilization constraints to achieve effective energy consumption control. CN110901710A discloses a train scheduling method based on dynamic passenger flow, which predicts the full load rate and transfer volume of line sections through real-time passenger flow data to dynamically adjust the train operation plan. However, the above-mentioned existing technologies mostly focus on single or local optimization objectives. For example, they either emphasize the utilization of regenerative braking energy or take passenger flow changes as the only optimization reference. They lack a comprehensive consideration of multi-objective coordinated optimization and are difficult to simultaneously take into account multiple dimensions such as train operation energy consumption, on-time performance, regenerative energy utilization efficiency, and power supply load management.

[0004] Specifically, existing urban rail transit operation scheduling methods have the following shortcomings: First, the optimization objectives are relatively singular or limited, ignoring the interrelationships between different objectives and failing to achieve overall optimal scheduling. Second, the collection, processing, and utilization of real-time data are insufficient, making it difficult to comprehensively and effectively reflect the dynamic characteristics of the rail transit system's operating status, thus limiting the accuracy of scheduling decisions. Third, in terms of solution methods, they rely heavily on traditional single-objective optimization or simple mathematical programming methods, resulting in insufficient computational efficiency and real-time performance, making it difficult to meet the actual needs of rapid response and efficient processing of large-scale data. In addition, although existing technical solutions propose regenerative braking energy utilization and dynamic passenger flow response mechanisms, they lack comprehensive consideration of important factors such as peak load and real-time electricity price fluctuations, leading to limitations in actual operation and failing to fully tap the potential for energy conservation and consumption reduction in rail transit.

[0005] Therefore, existing urban rail transit energy consumption scheduling technologies generally suffer from the problem of difficulty in simultaneously satisfying multi-objective coordinated optimization and real-time efficient decision-making. There is an urgent need for a comprehensive scheduling method that can take into account the reduction of traction energy consumption, full recovery and utilization of regenerative energy, guarantee of on-time operation, and reduction of power grid load. This invention effectively solves the problem of scheduling decision-making limitations caused by single optimization objectives in existing technologies by comprehensively processing real-time data, coordinating multi-objective optimization, and iterative solving with intelligent algorithms, and achieves optimal control of the overall operating energy consumption of the rail transit system. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] In view of the aforementioned existing problems, the present invention is proposed.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: synchronously collect train operation conditions, traction power supply parameters, passenger flow spatiotemporal distribution and time-of-use electricity price data on the train side, the track side and the power supply side, and sequentially perform cleaning, integration, conversion and reduction to output a standardized scheduling input dataset;

[0009] Using the dataset as input, an energy consumption scheduling model is established with the objectives of minimizing traction energy consumption, maximizing regenerative energy utilization, maximizing on-time operation rate, and minimizing peak load, and with speed curve, departure interval, and energy storage charging and discharging power as decision variables, constrained by train safety, line capacity, and power supply capacity.

[0010] The energy consumption scheduling model is iteratively solved using an artificial intelligence multi-objective optimization algorithm to obtain a set of candidate scheduling schemes that meet real-time constraints;

[0011] The target weight is adaptively determined based on the rolling time-domain evaluation function. The candidate scheduling scheme set is then comprehensively evaluated, the optimal scheduling scheme is selected, and the optimal scheme is sent to the vehicle control system in real time.

[0012] As a preferred embodiment of the urban rail transit energy consumption scheduling method based on artificial intelligence multi-objective optimization described in this invention, the output standardized scheduling input dataset includes:

[0013] The train's traction current, braking force, and speed curves are sampled at fixed intervals, and noise and null values ​​are removed using sliding window filtering and missing value interpolation.

[0014] Passenger flow density data is resampled according to a unified mileage and time index and aligned with train operation data in time and space.

[0015] The power supply monitoring interface is used to obtain feeder voltage, current and time-of-use electricity price, and these are synchronously mapped to the train data time axis.

[0016] Range normalization is performed on all numerical fields, one-hot encoding is performed on categorical fields, and principal component compression is used on high-dimensional continuous features to retain the main information.

[0017] After the data consistency verification is passed, the features of the train side, the line side and the power supply side are merged according to the sampling time, written into the distributed cache, and a standardized scheduling input dataset is output.

[0018] As a preferred embodiment of the urban rail transit energy consumption scheduling method based on multi-objective optimization using artificial intelligence as described in this invention, the establishment of the energy consumption scheduling model includes:

[0019] Traction energy consumption, regenerative energy utilization rate, on-time operation rate and peak load are set as independent optimization targets, and Pareto relationship is used to preserve the trade-off space of each target;

[0020] The train speed curve is discretized into segment target speed pairs, each of which is constrained by the line speed limit and the train acceleration capability.

[0021] The departure interval is used as an adjustable timetable parameter, with its lower limit being the sum of the interval travel time and the minimum safe tracking interval;

[0022] The energy storage charging and discharging power is limited to the range of the rated power capacity curve of the energy storage device and is associated with the regenerative braking energy recovery task.

[0023] Simultaneously, train tracking safety constraints, power supply system load margin constraints, and bus voltage stability constraints are introduced to form a multi-objective scheduling model description file for train-line-power supply coupling, which is the energy consumption scheduling model.

[0024] As a preferred embodiment of the urban rail transit energy consumption scheduling method based on artificial intelligence multi-objective optimization described in this invention, the energy consumption scheduling model is iteratively solved using an artificial intelligence multi-objective optimization algorithm, including:

[0025] In the offline phase, historical operational data is used to train a deep neural network predictor to estimate the values ​​of each objective function in milliseconds.

[0026] In the online phase, the predictor output is used as the fitness function, and the improved fast non-dominated sorting evolutionary algorithm is used to initialize and update the solution set.

[0027] Adaptive crossover and mutation are performed on each generation of the population, and an incremental evaluation strategy is used to calculate only the target value of the changed part, thus shortening the computation time.

[0028] When the non-dominated solution set is stable, output the Pareto front solution set that satisfies the real-time constraints, as a candidate scheduling scheme set.

[0029] As a preferred embodiment of the urban rail transit energy consumption scheduling method based on artificial intelligence multi-objective optimization described in this invention, the candidate scheduling scheme set specifically includes:

[0030] Target speed curve files corresponding to each train section;

[0031] A train departure plan document that clearly specifies the departure time and interval between trains within a section;

[0032] Time series file of charge and discharge power of energy storage devices used for energy recovery and peak electricity consumption reduction;

[0033] Each optimization objective value and constraint satisfaction mark is associated with each scheme.

[0034] As a preferred embodiment of the urban rail transit energy consumption scheduling method based on artificial intelligence multi-objective optimization described in this invention, the objective weights are adaptively determined according to a rolling time-domain evaluation function, including:

[0035] Real-time monitoring of passenger flow forecasting errors, energy price fluctuations, and power supply load factor within the rolling time domain, and generation of corresponding dynamic factors;

[0036] The basic weights for traction energy consumption, regenerative energy utilization rate, on-time operation rate, and peak load are set according to the operation priority table.

[0037] After multiplying the dynamic factor by the basic weight, the result is normalized so that the sum of all weights is 1.

[0038] When any indicator deviates beyond the threshold for two consecutive periods, an emergency weight correction strategy is triggered, increasing the weight of the target indicator while simultaneously decreasing the weights of other indicators.

[0039] As a preferred embodiment of the urban rail transit energy consumption scheduling method based on multi-objective optimization using artificial intelligence as described in this invention, the candidate scheduling scheme set is comprehensively evaluated according to the objective weights, including:

[0040] Calculate a weighted total score for each candidate solution based on the target weights;

[0041] The weighted scores are mapped to the 0-1 interval using a fuzzy satisfaction function, and then sorted by satisfaction level.

[0042] If the difference between the two most satisfactory options is less than a set threshold, then the renewable energy utilization rate is compared, and the option with the higher utilization rate is selected.

[0043] If it is still impossible to distinguish them, then the judgment will be carried out step by step according to the priority order of peak load, punctuality rate and traction energy consumption until a unique optimal solution is generated.

[0044] As a preferred embodiment of the urban rail transit energy consumption scheduling method based on multi-objective optimization using artificial intelligence as described in this invention, the optimal scheduling scheme is selected from the comprehensive evaluation results and sent to the vehicle control system in real time, including:

[0045] The candidate optimal solution is rapidly and safely verified on the digital twin simulation platform. The verification content includes at least train conflict check, power supply load margin verification and communication compatibility verification.

[0046] After verification, the optimal solution is translated into speed-time curve files, departure plan files, and energy storage control instruction files, and then distributed to the vehicle control system and power supply control center through the train communication network and power supply dispatching network, respectively.

[0047] The beneficial effects of this invention are as follows: By comprehensively applying multi-source data acquisition and processing technology, multi-objective optimization modeling method, artificial intelligence rapid iterative solution and rolling time domain evaluation strategy, this invention achieves comprehensive optimization of energy consumption scheduling in urban rail transit systems, effectively improves train operation energy efficiency, regenerative energy utilization rate and on-time operation rate, while reducing the peak load of traction power supply system, significantly improving the overall energy utilization efficiency and operation performance of rail transit, and achieving the dual goals of energy saving and consumption reduction and efficient operation of the system. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0049] Figure 1 This is a flowchart illustrating the urban rail energy consumption scheduling method based on multi-objective optimization using artificial intelligence, as shown in this invention. Detailed Implementation

[0050] 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. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0051] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0052] 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.

[0053] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a multi-objective optimization method for urban rail transit energy consumption scheduling based on artificial intelligence, which specifically includes the following steps:

[0054] S1. Simultaneously collect train operating conditions, traction power supply parameters, passenger flow spatiotemporal distribution, and time-of-use electricity price data from the train side, track side, and power supply side. Then, sequentially perform cleaning, integration, conversion, and reduction processes to output a standardized dispatch input dataset. Note that the following points should be noted in this step:

[0055] The train's traction current, braking force, and speed curves are sampled at fixed intervals, and noise and null values ​​are removed using sliding window filtering and missing value interpolation.

[0056] Passenger flow density data is resampled according to a unified mileage and time index and aligned with train operation data in time and space.

[0057] The power supply monitoring interface is used to obtain feeder voltage, current and time-of-use electricity price, and these are synchronously mapped to the train data time axis.

[0058] Range normalization is performed on all numerical fields, one-hot encoding is performed on categorical fields, and principal component compression is used on high-dimensional continuous features to retain the main information.

[0059] After the data consistency verification is passed, the features of the train side, the line side and the power supply side are merged according to the sampling time, written into the distributed cache, and a standardized scheduling input dataset is output.

[0060] As an example, data is collected from the train side:

[0061] Sensors: traction current, braking force, speedometer;

[0062] Sampling period: 50ms;

[0063] Data format: 32-bit floating point;

[0064] Value range: current 0~4kA, speed 0~120km / h.

[0065] As an example, data is collected from the line side:

[0066] Sensor: Video millimeter-wave passenger flow detection unit, outputting passenger flow density and platform identification (string);

[0067] Sampling period: 30s.

[0068] As an example, the power supply side data is collected as follows:

[0069] Interface: SCADA OPC-UA, fields must include at least feeder voltage, feeder current, and time-of-use electricity price;

[0070] Sampling period: 60s.

[0071] For example, data cleaning of the collected data includes: applying a 5-point sliding window mid-range filter to the traction current and speed curves; if the current value deviates from the mid-range of the window by ±3 times the root mean square, it is marked as abnormal and set as missing; if the length of consecutive missing values ​​is ≤3 sampling periods, linear interpolation is performed; if the length of consecutive missing values ​​is >3 sampling periods, forward padding is performed and "quality_flag" is marked as "LOW".

[0072] Furthermore, integrated processing is performed on the cleaned data, including:

[0073] Use Unix millisecond timestamps uniformly, and round all sampling records up to the nearest 10ms boundary;

[0074] The passenger flow density is mapped to the train mileage-time coordinate system according to the station number-platform mileage index, so as to achieve spatial alignment between train and passenger flow.

[0075] Furthermore, the integrated data undergoes data transformation, including:

[0076] Perform 0-1 range scaling on continuous fields such as traction current and speed;

[0077] One-hot encoding of discrete fields such as station number and power supply zone is used to generate Boolean vectors;

[0078] Finally, reduction is performed on the transformed data features, including: principal component compression is used on ≥50-dimensional working condition statistics vectors, retaining the first k dimensions (e.g., k=10) that explain ≥95% of the cumulative variance.

[0079] It should be noted that the data consistency verification in this embodiment includes checking the monotonicity of timestamps, the legality of the value range, and station consistency. If the record failure rate is >0.1%, an alarm is triggered and written to the audit log.

[0080] The final standardized scheduling input dataset is generated, represented in a structured format (Parquet), as shown in the following example:

[0081] {

[0082] "ts": int64, "train_id": string, "km_post": float,

[0083] "speed_norm": float, "traction_curr_norm": float,

[0084] "passenger_density":float,"price_norm":float,

[0085] "pca_feat_1"…"pca_feat_k":float,"quality_flag":string

[0086] }

[0087] The dataset is stored in the HDFS partition directory / rail / standardized / yyyyMMdd / HH / ;

[0088] It is also synchronously written to the Redis cache (key: ts|train_id, expires in 10 minutes) for subsequent models to call in real time.

[0089] It should be noted that this step effectively integrates multi-dimensional data such as train operation status, passenger flow trends, and power supply status to comprehensively reflect the real-time operation status of the urban rail transit system. This facilitates accurate modeling in subsequent optimization, ensures higher credibility and accuracy of data-driven decision-making, and improves the reliability of the data foundation for energy consumption scheduling decisions.

[0090] S2. Using the dataset as input, establish an energy consumption scheduling model with the objectives of minimizing traction energy consumption, maximizing regenerative energy utilization, maximizing on-time performance, and minimizing peak load. The model uses speed curves, departure intervals, and energy storage charging / discharging power as decision variables, and is constrained by train safety, line capacity, and power supply capacity. Note that the following should be noted in this step:

[0091] Traction energy consumption, regenerative energy utilization rate, on-time operation rate and peak load are set as independent optimization targets, and Pareto relationship is used to preserve the trade-off space of each target;

[0092] The train speed curve is discretized into segment target speed pairs, each of which is constrained by the line speed limit and the train acceleration capability.

[0093] The departure interval is used as an adjustable timetable parameter, with its lower limit being the sum of the interval travel time and the minimum safe tracking interval;

[0094] The energy storage charging and discharging power is limited to the range of the rated power capacity curve of the energy storage device and is associated with the regenerative braking energy recovery task.

[0095] Simultaneously, train tracking safety constraints, power supply system load margin constraints, and bus voltage stability constraints are introduced to form a multi-objective scheduling model description file for train-track-power supply coupling.

[0096] It should be noted that the aforementioned multi-objective scheduling model description file is written into a model description file that can be directly parsed by artificial intelligence multi-objective optimization algorithms, serving as the final expression of the energy consumption scheduling model.

[0097] As an example, the mathematical expression of the energy consumption scheduling model is as follows:

[0098]

[0099] Where u is the target speed decision vector for each sub-segment, v is the departure interval decision vector for each train segment, w is the discrete-time charging and discharging power vector of the energy storage device, t is the continuous time variable, T is the evaluation period length, α(t) is the time weighting function constructed based on the train load rate, and P tr (u,t) is the traction power curve obtained by velocity vector mapping, β(t) is the time weighting function constructed based on the regenerative energy absorption rate, and P re (w,t) represents the regenerative energy absorption power curve obtained from the energy storage power mapping, γ is the on-time performance sensitivity coefficient, M is the number of evaluation sections, and η k (v k (v) represents the departure interval. k The corresponding on-time satisfaction function, P gr (u,w,t) represents the instantaneous grid-side power curve after traction-energy storage coupling, P max Let F(·) be the upper limit of the allowed peak power of the power supply system, and let F(·) be a multi-objective function vector containing four components.

[0100] It should be further noted that in the comprehensive objective vector F(u,v,w)=(F1,F2,F3,F4), the four components correspond to the following optimization objectives in sequence:

[0101] Traction energy consumption target:

[0102]

[0103] Renewable energy utilization rate target:

[0104]

[0105] On-time performance target:

[0106]

[0107] Peak load target:

[0108]

[0109] After fractional normalization, each component of the comprehensive target vector F(u,v,w) belongs to the interval [0,1]. The smaller the value of the first component (F1), the lower the energy consumption per unit weighted time. The larger the value of the second component (F2), the higher the regenerative energy utilization rate. The closer the value of the third component (F3) is to 1, the higher the on-time rate. The smaller the value of the fourth component (F4), the lower the peak power load. When all four components reach the set threshold interval at the same time, it is considered as the optimal solution.

[0110] Preferably, this step achieves multi-objective collaborative optimization, systematically considers the inherent relationship between different operational objectives, enhances the systematicness and integrity of rail transit scheduling decisions, guides scheduling strategies to more accurately and efficiently meet diverse operational needs, thereby obtaining a globally optimized operational strategy, ensuring operational safety while achieving the goal of energy conservation and consumption reduction.

[0111] S3. Utilize an artificial intelligence multi-objective optimization algorithm to iteratively solve the energy consumption scheduling model, obtaining a set of candidate scheduling schemes that satisfy real-time constraints. Note the following in this step:

[0112] (1) Offline predictor training

[0113] Collect standardized scheduling input datasets and corresponding historical scheduling results from the past 12 months, and automatically align them with peak passenger flow, holidays, and abnormal operation segments;

[0114] A deep neural network containing temporal convolutional layers, bidirectional gated recurrent layers, and attention convergence layers is selected to perform multi-task learning on four targets: traction energy consumption, regenerative energy utilization rate, on-time operation rate, and peak load.

[0115] A hierarchical hold-out strategy is adopted to divide the training set, validation set and test set, and the model performance is jointly evaluated by the target mean squared error and Pareto ranking consistency rate.

[0116] When the average error of the four objectives is lower than the threshold set by the operator and the consistency rate is not lower than 0.9, the network parameters are frozen and exported as a "predictor-vN" model weight file for loading in the online phase.

[0117] (2) Online population initialization

[0118] At the beginning of each scheduling cycle, read a normalized input data fragment from the most recent hour as an environment state vector;

[0119] Based on the environmental state vector, the "predictor-vN" is invoked to quickly sample and estimate three types of variables: speed curve, departure interval, and energy storage power, forming 200 initial individuals;

[0120] The initial individuals are filtered by constraints on train safety, line capacity and power supply capacity. Solutions that do not meet the constraints are eliminated, and about 150 legal individuals are retained as population G0.

[0121] (3) Adaptive crossover-mutation and incremental evaluation

[0122] For population G t The system retrieves historical good solution fragments for similar environments from the operation log, compares them with the variance of the target component in the current population, and dynamically sets the crossover probability P. c With the probability of mutation P m (High variance → Low P) c High P m Low variance → High P c Low P m );

[0123] For new individuals generated by crossover mutation, only the time slice target affected by the changed variables is recalculated, and the unchanged part inherits the parent's estimate;

[0124] The new and old individuals are merged, and diversity is maintained using fast non-dominated sorting and crowding distance. The top 150 individuals are selected to generate the next generation G. {t+1} ;

[0125] (4) Convergence monitoring and real-time constraint judgment

[0126] The algorithm is considered to have converged when the similarity between two consecutive Pareto fronts is ≥0.95 and the total computation time has exceeded the scheduling cycle × 0.6.

[0127] Hard constraint verification and safety timing simulation are performed again on all individuals at the convergence front to eliminate schemes that have vehicle-following conflicts or voltage drops;

[0128] If there are fewer than 5 candidate solutions after elimination, reduce the similarity threshold and repeat steps (3) to (4) until the quantity requirement is met;

[0129] (5) Output the candidate scheduling scheme set

[0130] Three types of instruction files are generated for each candidate solution:

[0131] Target speed curve file: CSV format, with fields including section number, distance from station, and target speed;

[0132] Departure schedule file: JSON format, recording train ID, departure time, and interval between trains in the section;

[0133] Energy storage power sequence file: Parquet format, with fields including time stamp, power value, and operating mode identifier (charge / discharge / standby);

[0134] Calculate and attach the four objective standardized values ​​and the "constraint_pass" boolean flag, and encapsulate them uniformly as "Solution Package-UID";

[0135] All solution packages are stored in the scheduling buffer, and a "candidate solution ready" event is published through the message queue for the rolling evaluation module in step S4 to call.

[0136] As an example, the candidate solution set includes:

[0137] Scheme UID = A17F-3B2: Traction energy consumption 0.42, recycling 0.83, punctuality rate 0.91, peak load 0.35, constraint_pass = true;

[0138] Scheme UID = D04C-8E1: Traction energy consumption 0.40, recycling 0.78, punctuality rate 0.89, peak load 0.38, constraint_pass = true;

[0139] Scheme UID = F92B-5D0: Traction energy consumption 0.45, recycling 0.85, punctuality rate 0.93, peak load 0.33, constraint_pass = true.

[0140] Preferably, this embodiment utilizes artificial intelligence algorithms to search the optimal solution space more efficiently than traditional mathematical programming methods, and makes full use of historical data to train the artificial intelligence predictor, which greatly shortens the computation time, improves the real-time response capability of the algorithm and the feasibility of the scheduling scheme, and enables the solution-solving process to adapt to changes in operational status in real time, greatly enhancing the dynamic adaptability of scheduling decisions.

[0141] S4. Adaptively determine the target weight based on the rolling time-domain evaluation function, comprehensively evaluate the candidate scheduling scheme set, select the optimal scheduling scheme, and send it to the vehicle control system in real time. Note that the following points should be noted in this step:

[0142] Real-time monitoring of passenger flow forecasting errors, energy price fluctuations, and power supply load factor within the rolling time domain, and generation of corresponding dynamic factors;

[0143] The basic weights for traction energy consumption, regenerative energy utilization rate, on-time operation rate, and peak load are set according to the operation priority table.

[0144] After multiplying the dynamic factor by the basic weight, normalize the whole to make the sum of the weights equal to 1;

[0145] When any indicator deviates beyond the threshold for two consecutive periods, an emergency weight correction strategy is triggered to increase the weight of the target indicator and simultaneously decrease the weight of other indicators.

[0146] Calculate the weighted total score for each candidate solution based on the target weight;

[0147] The weighted scores are mapped to the 0-1 interval using a fuzzy satisfaction function, and then sorted by satisfaction level.

[0148] If the difference between the two most satisfactory options is less than a set threshold, then the renewable energy utilization rate is compared, and the option with the higher utilization rate is selected.

[0149] If it is still impossible to distinguish, then the decision will be made step by step according to the priority order of peak load, on-time rate, and traction energy consumption until a unique optimal solution is found.

[0150] The candidate optimal solution is rapidly and safely verified on the digital twin simulation platform. The verification content includes at least train conflict check, power supply load margin verification and communication compatibility verification.

[0151] After verification, the optimal solution is translated into speed-time curve files, departure plan files, and energy storage control instruction files, and then distributed to the vehicle control system and power supply control center through the train communication network and power supply dispatching network, respectively.

[0152] Preferably, by dynamically adjusting the evaluation weights, the present invention can respond promptly to changes in real-time operational conditions such as passenger flow fluctuations, electricity price changes, and load demand, ensuring the flexibility and adaptability of scheduling decisions, ensuring the accurate and efficient execution of scheduling decisions, and significantly improving the operational efficiency of the rail transit system.

[0153] The aforementioned method for performing rapid and secure verification of candidate optimal solutions on a digital twin simulation platform can be carried out using existing technologies and methods, and will not be elaborated further in this example.

[0154] 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 method for urban rail transit energy consumption scheduling based on multi-objective optimization using artificial intelligence, characterized in that, include: Simultaneously collect train operation status, traction power supply parameters, passenger flow spatiotemporal distribution and time-of-use electricity price data from the train side, track side and power supply side, and sequentially perform cleaning, integration, conversion and reduction to output a standardized scheduling input dataset; Using the dataset as input, an energy consumption scheduling model is established with the objectives of minimizing traction energy consumption, maximizing regenerative energy utilization, maximizing on-time operation rate, and minimizing peak load, and with speed curve, departure interval, and energy storage charging and discharging power as decision variables, constrained by train safety, line capacity, and power supply capacity. The establishment of the energy consumption scheduling model includes: setting traction energy consumption, regenerative energy utilization rate, on-time operation rate, and peak load as independent optimization objectives, and using Pareto relations to preserve the trade-off space of each objective; discretizing the train speed curve into segment target speed pairs, each pair being constrained by line speed limits and train acceleration capabilities; using the departure interval as an adjustable timetable parameter, with its lower limit being the sum of the segment running time and the minimum safe following interval; limiting the energy storage charging and discharging power within the range of the rated power capacity curve of the energy storage device, and associating it with the regenerative braking energy recovery task; and simultaneously introducing train following safety constraints, power supply system load margin constraints, and bus voltage stability constraints to form a multi-objective scheduling model description file for train-line-power supply coupling, which is the energy consumption scheduling model. The energy consumption scheduling model is iteratively solved using an artificial intelligence multi-objective optimization algorithm to obtain a set of candidate scheduling schemes that meet real-time constraints; The energy consumption scheduling model is iteratively solved, including: in the offline phase, historical operating data is used to train a deep neural network predictor to estimate the values ​​of each objective function in milliseconds; a deep neural network containing temporal convolutional layers, bidirectional gated recurrent layers, and attention convergence layers is selected to perform multi-task learning on four objectives: traction energy consumption, regenerative energy utilization rate, on-time operation rate, and peak load; a hierarchical hold-out strategy is adopted to divide the training set, validation set, and test set, and the model performance is jointly evaluated by the target mean square error and Pareto ranking consistency rate; in the online phase, the predictor output is used as the fitness function, and an improved fast non-dominated ranking is adopted. The evolutionary algorithm initializes and updates the solution set; at the beginning of each scheduling cycle, it reads a standardized input data fragment from the most recent hour as an environment state vector; based on the environment state vector, it calls the predictor -vN to quickly sample and estimate three types of variables: speed curve, departure interval, and energy storage power, forming 200 initial individuals; it applies constraints on train safety, line capacity, and power supply capacity to the initial individuals, eliminating solutions that do not meet the constraints and retaining 150 legal individuals as the population G0; it performs adaptive crossover and mutation on each generation of the population, and uses an incremental evaluation strategy to calculate only the target value of the changed part, shortening the computation time; it then applies the population G0... t The system retrieves historical good solution fragments for similar environments from the operation log, compares them with the variance of the target component in the current population, and dynamically sets the crossover probability P. c With the probability of mutation P m High variance → low P c High P m Low variance → High P c Low P m For new individuals generated by crossover mutation, only the time slice target affected by the changed variables is recalculated, and the unchanged part inherits the parent's estimate; when the non-dominated solution set is stable, that is, when the similarity of two consecutive Pareto fronts is ≥ 0.95 and the total computation time has exceeded the scheduling cycle × 0.6, the algorithm is considered to have converged; hard constraint verification and safe timing simulation are performed again on all individuals in the convergence front, and schemes that have following conflicts or voltage drops are eliminated; if there are less than 5 candidate solutions after elimination, the similarity threshold is reduced and the iteration is repeated until the quantity requirement is met; the Pareto front solution set that meets the real-time constraints is output as the candidate scheduling scheme set; The target weight is adaptively determined based on the rolling time-domain evaluation function, including: real-time monitoring of passenger flow forecasting errors, energy price fluctuations, and power supply load rate in the rolling time domain, and generating corresponding dynamic factors; setting basic weights for traction energy consumption, renewable energy utilization rate, on-time operation rate, and peak load based on the operation priority table; multiplying the dynamic factors with the basic weights and then normalizing the whole to make the sum of all weights equal to 1; when any indicator deviates from the threshold for two consecutive periods, an emergency weight correction strategy is triggered to increase the weight of the target indicator and simultaneously decrease the weights of other indicators. The candidate scheduling scheme set is comprehensively evaluated, including: calculating a weighted total score for each candidate scheme according to the target weight; mapping the weighted score to the 0~1 interval using a fuzzy satisfaction function and sorting them according to satisfaction; if the difference between the two schemes with the highest satisfaction is less than a set threshold, then the regenerative energy utilization rate is compared, and the one with the higher utilization rate is selected first; if they still cannot be distinguished, then they are judged step by step according to the priority order of peak load, on-time rate, and traction energy consumption until a unique optimal scheme is generated. The optimal scheduling scheme is selected and sent to the vehicle control system in real time. This includes: performing a rapid safety verification of the candidate optimal scheme on the digital twin simulation platform. The verification content includes at least inter-train conflict check, power supply load margin verification, and communication compatibility verification. After the verification is passed, the optimal scheme is translated into speed-time curve file, departure plan file, and energy storage control instruction file, and sent to the vehicle control system and power supply control center through the train communication network and power supply dispatching network, respectively.

2. The urban rail transit energy consumption scheduling method based on artificial intelligence multi-objective optimization according to claim 1, characterized in that, The output standardized scheduling input dataset includes: The train's traction current, braking force, and speed curves are sampled at fixed intervals, and noise and null values ​​are removed using sliding window filtering and missing value interpolation. Passenger flow density data is resampled according to a unified mileage and time index and aligned with train operation data in time and space. The power supply monitoring interface is used to obtain feeder voltage, current and time-of-use electricity price, and these are synchronously mapped to the train data time axis. Range normalization is performed on all numerical fields, one-hot encoding is performed on categorical fields, and principal component compression is used on high-dimensional continuous features to retain the main information. After the data consistency verification is passed, the features of the train side, the line side and the power supply side are merged according to the sampling time, written into the distributed cache, and a standardized scheduling input dataset is output.

3. The urban rail energy consumption scheduling method based on artificial intelligence multi-objective optimization according to claim 1, characterized in that, The candidate scheduling scheme set specifically includes: Target speed curve files corresponding to each train section; A train departure plan document that clearly specifies the departure time and interval between trains within a section; Time series file of charge and discharge power of energy storage devices used for energy recovery and peak electricity consumption reduction; Each optimization objective value and constraint satisfaction mark is associated with each scheme.

Images