Railway traction power supply system performance prediction energy management method and system based on macd momentum perception

Abstract

The application discloses a railway traction power supply system prediction performance energy management method and system based on MACD momentum perception. In view of the problems that the traditional prediction model has phase lag and prediction deviation when facing high-frequency fluctuation traction load, the application introduces the MACD index to extract the momentum characteristics such as the speed and acceleration of the load track, and uses the particle swarm optimization (PSO) to adaptively optimize the MACD parameters. Then the extracted momentum characteristics are fused to construct a deep learning prediction model, realizing high-precision advance prediction of the traction load. Further, a closed-loop rolling optimization framework is constructed, and the momentum perception prediction result is fed back to the energy management system (EMS) to guide the advance action of the roadside energy storage equipment. The application effectively eliminates the phase inconsistency between the control instruction and the actual demand, improves the recovery efficiency of the regenerative braking energy, and reduces the system operation cost.

Description

Technical Field

[0001] This invention relates to the fields of electrical engineering and power system automation and energy dispatching technology, and in particular to a predictive energy management method and system for railway traction power supply system based on MACD momentum sensing. Background Technology

[0002] The rapid development of high-speed electrified railways is an important part of carbon reduction in the transportation sector. However, traction loads are characterized by high randomness and drastic fluctuations, posing a significant challenge to the stable operation of the power system. To improve power quality and recover regenerative braking energy, railway power conditioners (RPCs) and roadside energy storage systems (ESSs) are typically deployed in the traction power supply system.

[0003] In actual operation, traditional control strategies suffer from response delays when faced with rapid load shifts (such as simultaneous acceleration or braking of multiple trains), easily missing the optimal opportunity for energy recovery. To support the forward-looking scheduling of roadside energy storage systems, high-precision Short-Term Load Forecasting (STLF) is essential. Although deep learning models such as Long Short-Term Memory (LSTM) networks and Temporal Convolutional Networks (TCN) are widely used, due to the rapid ramp-up characteristics of traction loads, existing models often only act as "low-pass filters" during prediction, ignoring the evolution of higher-order dynamic information of the load, exhibiting a serious "phase inconsistency" phenomenon (i.e., phase lag or advance). This phase misalignment will lead to incorrect instructions issued by the downstream energy management system (EMS), causing the energy storage system to discharge too late during critical peak periods and charge untimely during braking periods, reducing the economy and reliability of energy dispatch. Summary of the Invention

[0004] To address the issue of phase inconsistency in traditional deep learning prediction models, which leads to delayed scheduling commands, low overall economic efficiency, and low energy recovery rate in roadside energy storage systems, this invention provides a predictive energy management method and system for railway traction power supply systems based on MACD momentum sensing.

[0005] The technical solution adopted in this invention is:

[0006] A predictive energy management method for railway traction power supply systems based on MACD momentum sensing includes the following steps:

[0007] Step 1: Obtain historical traction load data from the traction substation as the raw time series data.

[0008] Step 2 employs the Particle Swarm Optimization (PSO) algorithm to adaptively seek the optimal configuration of the exponentially smoothed moving average of difference (MACD) parameters with the goal of maximizing feature information entropy.

[0009] Step 3: Based on the optimal MACD parameters, extract the momentum feature sequence from the original time series data that can characterize the "velocity" and "acceleration" of the traction load trajectory.

[0010] Step 4 involves splicing and fusing the extracted continuous multi-step momentum feature sequence with the original historical traction load to construct and train a high-precision traction load prediction model based on deep learning to predict future load trajectories.

[0011] Step 5 inputs the future load forecast trajectory into the Model Predictive Control (MPC) framework, establishes the objective function by jointly optimizing time-of-use pricing and the state of charge of the energy storage system, and performs rolling optimization and energy dispatch execution.

[0012] The beneficial effects of this invention are as follows:

[0013] This invention explicitly introduces the MACD indicator, originally used in the financial field, to analyze the inertial correlation dynamics of train groups, successfully extracting the speed (current trend) and acceleration (impulse intensity) characteristics of the traction load. This overcomes the limitation of traditional deep learning, which relies solely on zero-order smoothing, reducing the short-term prediction root mean square error (RMSE) by up to 24.18% and effectively eliminating phase lag in system predictions. Based on this, a closed-loop rolling predictive energy scheduling framework is developed. Quantitative experiments show that this approach can achieve a 7.89% reduction in operating costs and a 39.05% increase in peak shaving benefits, demonstrating significant potential for practical engineering applications and economic value. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall process of a railway predictive energy management method based on MACD momentum sensing, provided in an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram comparing the inconsistency in the predicted phase of the traditional model and the momentum sensing model for rapid uphill loads in an embodiment of the present invention.

[0016] Figure 3 The graph shows a comparison of the prediction results of the momentum sensing model and the traditional model in this embodiment of the invention at different prediction step sizes (h=1, 5, 30, 60).

[0017] Figure 4 This is a schematic diagram of the overall energy storage optimization results when the prediction step size is 30 minutes in an embodiment of the present invention (including local magnified analysis window markings).

[0018] Figure 5 This is a comparative evaluation chart of the grid-connected power smoothing effect when the prediction step size is 30 minutes in an embodiment of the present invention.

[0019] Figure 6 This is a detailed analysis chart within a representative time window (prediction step size 30 minutes, t=802-921 minutes) in an embodiment of the present invention; wherein, Figure 6 (a) Comparison of load forecasts Figure 6 (b) Comparison of power dispatching for energy storage systems Figure 6 (c) For grid-connected power smoothing effect Figure 6 (d) is the trajectory of the state of charge (SOC). Detailed Implementation

[0020] The steps of the present invention will be described in detail below with reference to specific embodiments.

[0021] Step 1: Obtain high-resolution raw traction historical load data Based on Taylor expansion analysis, the mathematical characteristics of future load moments are determined:

[0022] (1)

[0023] in The "rate" of change in current load. This represents the "acceleration" of the current load change. Traditional models are typically equivalent to... This results in the loss of higher-order terms, leading to phase inconsistency.

[0024] Step 2: Construct a Particle Swarm Optimization (PSO) module to adaptively solve for the optimal computational parameter configuration for MACD. .

[0025] Design a fitness function We encourage the extraction of momentum features that exhibit strong volatility and high separation:

[0026] (2)

[0027] in, and These represent the standard deviation and the mean absolute value of the MACD histogram sequence, respectively. It is the standard deviation of the original load sequence; It is the correlation coefficient between the difference value (DIF) and the signal line (DEA). The optimization objective is to minimize... To extract the best trend reversal signals.

[0028] Step 3: Calculate the MACD momentum characteristic sequence based on the parameters output by PSO. The MACD calculation formula is as follows:

[0029] 1) Calculate the long and short period exponential moving averages (EMA):

[0030] (3)

[0031] The short period is derived from this formula. and long cycle .

[0032] 2) Calculate the difference value (DIF) to represent the first-order momentum (velocity):

[0033] (4)

[0034] 3) Calculate the difference average (DEA) to characterize trend smoothing:

[0035] (5)

[0036] 4) Calculate the MACD histogram ( Characterizing second-order momentum (acceleration):

[0037] (6)

[0038] Step 4: Compare the historical load sequence with the data calculated in Step 3. Feature concatenation is performed and the data is input into a deep learning prediction network with a VMD-TCN-LSTM architecture. The model overcomes its dependence on a single image by perceiving higher-order derivative information, thereby outputting a future prediction sequence that synchronously approximates the actual trajectory. .

[0039] Step 5: Design a rolling time-domain optimized energy scheduling strategy. Establish an optimization objective function to minimize the daily integrated operating cost. :

[0040] (7)

[0041] in The cost of purchasing electricity from the grid is determined by the real-time grid power. With time-of-use electricity pricing Multiplying them together yields the result; Costs associated with the degradation of energy storage equipment; Costs are levied for capacity penalties; To generate revenue from peak shaving and valley filling.

[0042] The following system operation constraints should be followed when minimizing the above objectives:

[0043] Power balance constraints:

[0044] (8)

[0045] State of Charge (SOC) update constraints:

[0046] (9)

[0047] Within each scrolling control window (e.g., 1 minute), only the first step of the charge / discharge command is issued. This allows the energy storage converter to operate, achieving the goal of predictive regulation before load peaks or severe braking, thus completely alleviating control delays and improving the regenerative braking energy recovery rate.

[0048] The method provided in this embodiment of the invention is not limited to the specific combination of the above steps. Based on the above teachings, those skilled in the art can implement various possible variations of the invention.

Claims

1. A predictive energy management method for railway traction power supply systems based on MACD momentum sensing, characterized in that, Includes the following steps: Obtain the original historical traction load data of the traction substation; A parameter optimization module based on particle swarm optimization algorithm is established. The fitness function is to maximize the feature information entropy and feature separation degree to adaptively optimize the parameters of the exponential smoothed moving average of differences algorithm and determine the optimal short period parameters, long period parameters and signal period parameters. Based on the optimal short-period parameters, long-period parameters, and signal period parameters, the momentum characteristic sequence of the original historical traction load data is calculated. The momentum characteristic sequence includes a first-order momentum characteristic for characterizing the load change rate and a second-order momentum characteristic for characterizing the load change acceleration. The momentum feature sequence is concatenated with the original historical traction load data and input into a pre-trained deep learning prediction network to obtain the traction load prediction trajectory for future time periods. The predicted traction load trajectory is input into the model predictive control framework. A rolling optimization model with the goal of minimizing the overall operating cost is established by combining the operating constraints of the roadside energy storage system. The optimal charging and discharging power control command of the roadside energy storage system is obtained and issued within the current control cycle.

2. The predictive energy management method for railway traction power supply system based on MACD momentum sensing according to claim 1, characterized in that, The calculation of the momentum characteristic sequence of the original historical traction load data includes: Calculate the current time Exponential moving average over period $N$ : in, This represents the actual traction load at the current moment. Based on the optimal short-period parameters and long-period parameters Calculate the difference value Let this be the first-order momentum characteristic: Based on the optimal signal period parameters Calculate the average line of the difference value : Calculate the difference between the two to obtain the histogram sequence. As the second-order momentum characteristic:

3. The predictive energy management method for railway traction power supply system based on MACD momentum sensing according to claim 2, characterized in that, Fitness function designed for the particle swarm optimization algorithm for: in, To find the optimal parameter combination For bar chart sequence standard deviation The mean of the absolute values ​​of the histogram sequence. The standard deviation of the original historical traction load sequence. It is a very small constant. The Pearson correlation coefficient is the sum of the differences and the mean of the differences. and As a penalty weighting coefficient, This is a penalty term constraining the smoothness of the signal period.

4. The predictive energy management method for railway traction power supply system based on MACD momentum sensing according to claim 1, characterized in that, The deep learning prediction network adopts a hybrid architecture that cascades variational mode decomposition, temporal convolutional network and long short-term memory network. Specifically, variational mode decomposition is first used to decompose the input concatenated features into multiple modal components. Then, local temporal features are extracted by temporal convolutional network and long short-term memory network is used to model long-term temporal dependencies. Finally, the total prediction load is aggregated and output through fully connected layer.

5. A predictive energy management method for railway traction power supply system based on MACD momentum sensing according to claim 1, characterized in that, The objective function used to establish a rolling optimization model aimed at minimizing overall operating costs. for: in, To optimize the time window, For the cost of purchasing electricity from the power grid, For the degradation cost of energy storage systems, The capacity penalty cost for power imbalance, For the subsidy revenue generated from peak shaving and valley filling; Furthermore, the following operational constraints must be satisfied during the process of solving the rolling optimization model: the total exchange power between the energy storage system and the power grid at the current node is equal to the predicted traction load at this moment, and the state of charge update of the energy storage system between adjacent moments follows the equation constraint of the physical laws of charging and discharging efficiency and capacity decay.

6. A predictive energy management system for railway traction power supply based on MACD momentum sensing, characterized in that, include: The data acquisition module is used to acquire the original historical traction load data of the traction substation; The parameter adaptive optimization module is used to adaptively optimize the parameters of the exponential smoothed moving average of difference algorithm using the particle swarm optimization algorithm to determine the optimal short-period parameters, long-period parameters, and signal period parameters. The momentum feature extraction module is used to calculate the momentum feature sequence of first-order velocity and second-order acceleration, which characterize the high-order time dynamic changes of the original traction load, based on the optimal parameters. The load rolling prediction module is used to fuse the momentum feature sequence with the original historical traction load data and input it into a deep learning network to output a forward-looking future traction load prediction trajectory that eliminates phase lag. The energy rolling distribution module is used to solve the energy storage system action sequence that minimizes operating costs under the model predictive control framework based on the traction load predicted trajectory, and to issue charging and discharging scheduling instructions for the current moment in real time.

7. 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 method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images