Intelligent scoring and self-learning updating method based on railway vehicle-mounted phased array antenna

Abstract

The application discloses a kind of intelligent scoring and self-learning updating method based on railway vehicle-mounted phased array antenna, comprising: data acquisition, comprehensive score generation, parameter configuration, credibility evaluation and self-learning update, signal switch and beam switching decision, data reporting.Comprehensive score introduces eight-dimensional parameter weighted calculation of RSRP, SINR, distance, speed, path loss, historical handover success rate, beam residence time, CQI, etc.The weight is dynamically configured by three methods of scene pre-configuration, historical data statistics self-learning and real-time feedback closed-loop adjustment.Introduce time attenuation factor to reduce the weight of old data.Credibility evaluation calculates comprehensive credibility from GPS accuracy, speed stability and sampling continuity.Three-dimensional.The self-learning update adopts hierarchical correction strategy.The application solves the problems of single signal evaluation dimension, lack of data credibility and inability to handle deviation hierarchically, and realizes predictive intelligent control.

Description

Technical Field

[0001] This invention relates to the field of railway mobile communication antenna control technology, specifically a smart scoring and self-learning update method based on railway vehicle-mounted phased array antennas. Background Technology

[0002] When high-speed trains operate at speeds exceeding 300 km / h, the train-to-ground communication link faces severe challenges such as rapid signal changes, multipath fading, and frequent switching. Phased array antennas, due to their electrically adjustable beams and fast response, have gradually become the preferred solution for high-speed train onboard antennas. However, how to achieve intelligent control of the phased array antenna beams, enabling it to dynamically optimize beam pointing based on multi-dimensional information such as train position and signal quality, remains a pressing technical challenge.

[0003] In existing technologies, antenna control systems typically employ the following decision-making methods: Reactive handover based on real-time signal strength: This method compares the current received power (RSRP) or signal-to-noise ratio (SINR) with a preset threshold, triggering handover when the signal quality falls below the threshold. This approach suffers from significant lag, often leading to communication interruptions in high-speed mobile scenarios due to delayed handover.

[0004] Single-dimensional signal evaluation: Considering only RSRP or SINR as a single indicator cannot comprehensively reflect signal quality and communication link adaptability. For example, even if RSRP is good but SINR is poor, the actual communication quality is still poor.

[0005] Lack of self-learning ability: It cannot use historical data to optimize decisions, and it needs to be re-scanned and evaluated every time it passes the same location, making it unable to achieve predictive switching.

[0006] Lack of hierarchical processing mechanism: The signal fluctuations lack fine-grained processing and cannot distinguish between abnormal fluctuations, environmental changes and normal fluctuations, resulting in database instability or delayed updates.

[0007] Lack of data credibility: In actual operation, factors such as GPS positioning errors, drastic fluctuations in vehicle speed, and uneven sampling intervals can lead to inconsistent data quality. Existing technologies do not assess data credibility, and using low-quality data directly for self-learning will pollute the database and reduce the accuracy of decision-making.

[0008] Therefore, how to construct an intelligent digital processing method that can integrate multi-dimensional signal parameters, introduce credibility assessment, possess self-learning capabilities, and handle signal deviations in a hierarchical manner is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0009] To address the technical problems of existing technologies, such as single signal evaluation dimensions, lack of self-learning ability, inability to grade signal deviations, and lack of data reliability, this invention provides an intelligent scoring and self-learning update method based on railway vehicle-mounted phased array antennas. This method enables predictive intelligent control of phased array antennas and is particularly suitable for beam selection and signal switching control of phased array antennas in high-speed railway scenarios.

[0010] To achieve the above objectives, the present invention provides the following technical solution: A smart scoring and self-learning update method based on a railway vehicle-mounted phased array antenna includes the following steps: Step S1: Data Acquisition: Acquire application scenario data, including device location information, reference signal received power (RSRP) of each beam, signal-to-noise ratio (SINR) of each beam, transmit power of each beam, current train speed, downlink channel quality indicator (CQI), and historical handover success rate; among which, historical trajectory data can be manually imported as the initial data source and continuously corrected through self-learning during device operation. Step S2: Initial Comprehensive Score Generation: Based on the application scenario data obtained in Step S1, an initial historical comprehensive score is generated; for phased array antenna applications, comprehensive scores for each channel under different beams are generated. The formula for calculating the overall score S is: S_Real = W1×F1(RSRP) + W2×F2(SINR) + W3×F3(Distance) + W4×F4(Speed) + W5×F5(Path Loss) + W6×F6(Handover Success Rate) + W7×F7(Dwell Time) + W8×F8(CQI); The functions are defined as follows: F1(RSRP) - RSRP normalization function, F1(RSRP)=(RSRP-RSRP_min) / (RSRP_max-RSRP_min), RSRP_min=-120dBm, RSRP_max=-60dBm.

[0011] F1(SINR) - SINR normalization function F2(SINR)=(SINR-SINR_min) / (SINR_max-SINR_min), SINR_min=-5dB, SINR_max=30dB.

[0012] F3 (Distance) - Distance-related function F3 (distance) = 1 - (d / d_max), d represents the current distance from the base station. d_max is the coverage radius of the base station.

[0013] F4 (Speed) - Speed-related functions F4 (speed) = 1 - (v / v_max) v represents the current speed. v_max=450km / h (can be set).

[0014] F5 (Path Loss) - Path loss function Path loss PL = Transmit power - RSRP (in dB), normalization function: F5(PL)=(PL_max-PL) / (PL_max-PL_min), PL_min=80dB PL_max=120dB The lower the path loss, the higher the score.

[0015] F6 (Handover Success Rate) - Historical handover success rate function. F6(HOSR) = HOSR, which is the historical success probability of switching at this location / beam. The value ranges from 0 to 1 and is obtained from historical data statistics.

[0016] F7 (Dwell Time) - Beam dwell time function. F7(Dwell)=min(Dwell / Dwell_max,1), where Dwell is the expected settling time (in seconds) of the beam at the current position, and Dwell_max=10 seconds. The longer the settling time, the higher the score.

[0017] F8 (CQI) - Channel Quality Indication Function F8(CQI)=(CQI-CQI_min) / (CQI_max-CQI_min), CQI_min=0, CQI_max=15.

[0018] Dynamic configuration of weighting coefficients: The weight coefficients W1 to W8 satisfy ∑Wi=1, and their dynamic configuration methods include: Scene-specific pre-configuration based on route characteristics: Different weight combinations are pre-configured for different route scenarios such as tunnels, viaducts, plains, mountains, and urban areas, and automatically switched through route recognition; Statistical self-learning based on historical data: Statistically analyze the variance distribution of each parameter in historical operation, and reduce the weight of a parameter when its variance is significantly greater than that of other parameters, and increase the weight of a parameter when its variance is significantly less than that of other parameters. Closed-loop adjustment based on real-time feedback: The actual communication quality after beam switching is used as feedback, and the weights are fine-tuned by gradient descent.

[0019] Step S3: Parameter Configuration: Configure the calculation parameters for the comprehensive score, including the weights of each parameter W1~W8, the intelligent switch threshold S_th, the statistical duration T, the threshold duration percentage P, and the time constant τ; the configuration parameters support local configuration and remote configuration by the network management system.

[0020] Step S4: Credibility Assessment and Self-Learning Update This step includes two sub-parts: credibility assessment and self-learning update.

[0021] S4.1 Credibility Assessment For each sampled data point, a confidence weight C is calculated, and the following three factors are comprehensively evaluated: GPS signal quality reliability C_GPS: calculated based on the GPS positioning accuracy factor (DOP). When DOP < 2, the positioning accuracy is high, and C_GPS = 1.0; When 2 ≤ DOP < 4, the positioning accuracy is moderate, and C_GPS = 0.7; When DOP≥4, the positioning accuracy is low, C_GPS=0.4; If there is no GPS signal, C_GPS=0.

[0022] Vehicle speed change rate confidence C_speed: calculated based on the vehicle speed change rate |Δv / Δt| compared to the last sample. When |Δv / Δt| < 2 m / s², the vehicle speed changes continuously, and C_cont = 1.0; When 2 ≤ |Δv / Δt| < 4 m / s², the vehicle speed changes slightly more, and C_cont = 0.6; When |Δv / Δt|≥4m / s², the vehicle speed changes too much, and C_cont=0.3.

[0023] Sampling continuity confidence C_cont: calculated based on the time interval Δt_samp from the last sample. When Δt_samp < 1 second, sampling is continuous, and C_cont = 1.0; When 1 ≤ Δt_samp < 2 seconds, the sampling interval is slightly larger, and C_cont = 0.8; When Δt_samp ≥ 2 seconds, the sampling interval is too large, and C_cont = 0.5. The overall credibility weight C = C_GPS × C_speed × C_cont.

[0024] When C < 0.3, the reliability of the sampled data is deemed too low, and it is discarded directly without entering the self-learning update process.

[0025] S4.2 Self-Learning Update: Using location information as a label, it records the historical integrated field strength information and phased array antenna beam switching status corresponding to that point, including: Calculate the closest point from the real-time location to the historical trajectory, and considering the continuity of time, use this point as a comparison reference point for the real-time location; Calculate the comprehensive score S_real and S_hist of the real-time location point = [Σλ(t_i)×C_i×S_i] / Σ(λ(t_i)×C_i) (using time decay weighted average), where C_i is the confidence weight of each historical sample; Historical data is subject to time decay when used in calculations. The weight of historical data at the current time t is λ(t_i) = e^(-(t-t_i) / τ), where t_i is the data recording time and τ is a time constant (default 7 days). Older data has lower weight, making the database focus more on recent changes. Calculate the comprehensive score S_real for the real-time location point and compare it with S_hist to obtain the deviation percentage ∆S = |S_real - S_hist| / S_hist × 100%; Implement a graded correction strategy based on the magnitude of the deviation: When the short-term real-time S deviation exceeds 30%, it is treated as an outlier, no correction is made, and historical S is used for comparison. When the long-term real-time S deviation exceeds 30%, the recorded value is corrected based on the deviation magnitude, and the real-time S is used for comparison. When the real-time S deviation is between 10% and 30%, the recorded value is corrected based on the deviation magnitude, and the real-time S is used for comparison. When the real-time S deviation is between 0 and 10%, it is considered a normal fluctuation and no correction is made; the real-time S is used for comparison. When updating the database, the contribution of new data is weighted according to its credibility weight C, that is: new record value = (old weighted average × total weight + C × S_real) / (total weight + C).

[0026] Step S5: Signal switching and beam switching decision: Compare the real-time comprehensive score S with the set threshold S_th: Within the statistical duration T, when the percentage of time below the threshold reaches P, a repeater signal activation command is output. When the percentage of time exceeding the threshold reaches P within the statistical duration T, a repeater signal shutdown command is output. For phased array antenna applications, the real-time S of the current beam is compared with the historical S of other beams (weighted by time attenuation), the beam with the highest score is selected as the target beam, and a beam switching command is output. Step S6: Data Reporting: Regularly report the smart switch data to the network management system. The reported data includes: data recording time, location information, field strength information, comprehensive score S automatic correction record, switch switching record, and beam switching status.

[0027] The beneficial effects of this invention are as follows: 1. Multi-dimensional Parameter Weighted Comprehensive Scoring System: This invention integrates eight dimensions of parameters, including RSRP, SINR, distance, speed, path loss, historical handover success rate, beam dwell time, and CQI, into a more comprehensive signal quality evaluation system. A time decay factor is introduced, causing the weight of historical data to decrease over time, making the database more focused on recent changes and adapting to the dynamic evolution of the line environment. Weights are dynamically configured through three methods: scenario-specific pre-configuration, historical data statistical self-learning, and real-time feedback closed-loop adjustment, solving the practical problem of determining weights.

[0028] 2. Graded Deviation Correction Mechanism: This invention is the first to apply a graded deviation amplitude processing strategy to an antenna signal evaluation system. Signal fluctuations are categorized into four types (outliers, long-term deviations, moderate deviations, and normal fluctuations), and completely different processing logics are applied to each type—outliers do not pollute the database, updates are sensitive to environmental changes, and normal fluctuations remain stable. This graded strategy achieves a balance between database stability and sensitivity, solving the technical challenge of simultaneously ensuring stability and response speed in traditional methods. The correction formula is weighted by confidence level C: S = (C_hist×S_hist+ C×S_Real) / (C_hist+C), taking into account both historical weights and real-time trends. The introduction of the confidence level assessment mechanism further improves data quality.

[0029] 3. Reliability Assessment and Weighted Update: This invention introduces a data reliability assessment mechanism for the first time in railway vehicle-mounted antenna signal processing methods. It comprehensively evaluates the reliability weight of each sample from three dimensions: GPS accuracy, vehicle speed stability, and sampling continuity. Low-reliability data is discarded, while high-reliability data contributes a greater weight in database updates. This mechanism effectively prevents low-quality data from polluting the database due to positioning errors, drastic speed changes, uneven sampling, etc., significantly improving the robustness and accuracy of the self-learning process.

[0030] 4. Self-learning and evolution capabilities: Through the accumulation of historical data and time decay weighted average updates, the database is continuously optimized as the number of train runs increases, achieving an adaptive capability that becomes more accurate with use.

[0031] 5. Predictive intelligent decision-making: Based on historical data from location tags, the system can predict the signal coverage ahead and generate beam switching commands in advance, achieving zero-delay, seamless switching and solving the lag problem of traditional reactive switching in high-speed scenarios.

[0032] 6. Dual-mode decision support: It supports two application scenarios simultaneously: repeater signal switching control and phased array antenna beam switching, and has good versatility and scalability.

[0033] 7. Complete data traceability capability: Records key information such as the comprehensive score correction process and switch records, which facilitates network management monitoring and subsequent analysis, and provides data support for system optimization. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in 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.

[0035] Figure 1 This is an overall flowchart of the intelligent comprehensive scoring and self-learning digital processing method according to an embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram illustrating the comprehensive scoring calculation and parameter configuration of the present invention.

[0037] Figure 3 This is a flowchart of the credibility assessment and self-learning update process of the present invention.

[0038] Figure 4 This is a flowchart illustrating the signal switching and beam switching decision-making process of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0041] Example 1: Overall Process: like Figure 1 As shown, the present invention provides an intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna, comprising the following steps: Step S1: Data Acquisition Data acquisition supports both manual import and real-time acquisition. Manual import is used for system initialization and must include data from at least three complete running scenarios. Real-time acquired data includes: GPS latitude and longitude, speed, RSRP, SINR, CQI, and transmit power.

[0042] Step S2: Initial overall score generation: like Figure 2 As shown, the overall score S is calculated using an eight-parameter weighted model. The following is a specific calculation example: The train collected data at the tunnel entrance: RSRP=-85dBm, SINR=12dB, distance from the base station d=300m, speed v=300km / h, transmit power=23dBm, historical handover success rate of this beam=0.95, dwell time=8 seconds, CQI=10.

[0043] The calculation methods for each parameter are as follows: F1 = (-85+120) / ( -60+120) = 35 / 60 = 0.583; F2 = (12+5) / (30+5) = 17 / 35 = 0.486 (SINR_min=-5dB); F3 = 1 - 300 / 500 = 0.4; F4 = 1 - 300 / 350 = 0.143; PL = 23 - (-85) = 108dB, F5 = (120-108) / (120-80) = 12 / 40 = 0.3; F6 = 0.95; F7 = min(8 / 10,1) = 0.8; F8 = (10-0) / (15-0) = 0.667; The tunnel scene weights are: W1=0.15, W2=0.25, W3=0.10, W4=0.05, W5=0.10, W6=0.15, W7=0.10, W8=0.10.

[0044] S = 0.15×0.583 + 0.25×0.486 + 0.10×0.4 + 0.05×0.143 + 0.10×0.3 +0.15×0.95 + 0.10×0.8 + 0.10×0.667 = 0.0875 + 0.1215 + 0.04 + 0.0072 + 0.03 + 0.1425 + 0.08 + 0.0667 = 0.5754, which is approximately 57.5 on a percentage scale.

[0045] tunnel 0.15 0.25 0.10 0.05 0.10 0.15 0.10 0.10 elevated road 0.20 0.20 0.15 0.05 0.10 0.10 0.10 0.10 Plains 0.25 0.15 0.15 0.05 0.10 0.10 0.10 0.10 mountainous areas 0.15 0.25 0.15 0.05 0.10 0.10 0.10 0.10 urban area 0.15 0.15 0.20 0.05 0.10 0.15 0.10 0.10 Step S3: Parameter Configuration: This step provides the following configurable parameters, supporting both local and remote configuration by the network management system: Weight parameters W1~W8 Percentage of the overall score Refer to step S2 Smart switch threshold S_th Scoring threshold for judging signal quality 70 Statistical duration T Time window for threshold determination 10 seconds Threshold duration percentage P The threshold for the duration of the trigger switch action 60% The configuration parameters simultaneously record data such as the current beam and real-time S, which can be reported to the network management system for monitoring.

[0046] Step S4: Credibility Assessment and Self-Learning Update like Figure 3 As shown, this step details the credibility assessment and self-learning update process.

[0047] Sub-step S4.1: Data Tagging: Using location information as a label, record the historical comprehensive field strength information corresponding to that point, including: Physical Cell Identifier (PCI); Reference signal received power (RSRP) for each beam; Signal-to-noise ratio (SINR) of each beam; Phased array antenna beam switching status; Sampling timestamp.

[0048] Sub-step S4.2: Reference point matching: Calculate the nearest point from the real-time location to the historical trajectory, and considering temporal continuity (i.e., the positional continuity between adjacent time points), use this point as the comparison reference point for the real-time location. The matching algorithm is as follows: Calculate the Euclidean distance between the real-time location (Lon_cur, Lat_cur) and each point (Lon_i, Lat_i) on the historical trajectory; Select the point with the smallest distance as the candidate reference point; Check the continuity between the timestamp of the candidate point and the current time. If the time jump is too large, interpolate by combining it with the second nearest point.

[0049] Sub-step S4.3: Credibility assessment: Assume that at the time of sampling, the GPS DOP = 1.5, the vehicle speed change rate = 0.5 m / s², and the interval from the last sampling = 0.8 seconds; C_GPS = 1.0 (DOP<2); C_speed = 1.0 (rate of change < 2 m / s²); C_cont = 1.0 (interval < 1 second); The overall credibility C = 1.0 × 1.0 × 1.0 = 1.0, indicating high data credibility. The system then proceeds to the self-learning process.

[0050] If the GPS signal is lost inside the tunnel (C_GPS=0), then C=0, the sampled data is discarded and will not be used for database updates.

[0051] Sub-step S4.4: Hierarchical correction processing: Different correction strategies are adopted based on the magnitude of the deviation and the execution: ∆S>30% and short duration Abnormal interference (such as momentary occlusion) Discard the outlier without making any corrections. Using historical S ∆S>30% and for a long time Environmental changes (such as new buildings or base station adjustments) Correcting records, using real-time S Using real-time S 10%<∆S ≤ 30% Obvious changes (such as seasonal vegetation changes) Correcting records, using real-time S Using real-time S 0 ≤ ∆S ≤ 10% Normal fluctuations (such as multipath effects) No modifications will be made; the system will remain stable. Using real-time S "Short time" refers to a few samplings with deviations, while "long time" refers to multiple consecutive samplings (e.g., more than 5 consecutive samplings) with deviations.

[0052] Sub-step S4.5: Self-learning update: Assume there are 5 historical records at point A at the tunnel entrance. The time, score, and reliability of each record are shown in the table below: 1 10 days 75 0.9 0.24 75×0.9×0.24=16.2 2 7 days 78 1.0 0.37 78×1.0×0.37=28.9 3 5 days 80 0.8 0.49 80×0.8×0.49=31.4 4 2 days 82 1.0 0.75 82×1.0×0.75=61.5 5 1 day 85 1.0 0.87 85×1.0×0.87=74.0 Total weighted denominator = 0.9×0.24 + 1.0×0.37 + 0.8×0.49 + 1.0×0.75 + 1.0×0.87 = 0.216 + 0.37 + 0.392 + 0.75 + 0.87 = 2.598; Weighted numerator = 16.2 + 28.9 + 31.4 + 61.5 + 74.0 = 212.0; S_hist = 212.0 / 2.598 ≈ 81.6; Current real-time sampling: S_real=70, deviation ∆S = |70-81.6| / 81.6 = 14.2%, which falls within the "10%~30%" range and is judged as a moderate change. Database correction is needed.

[0053] When updating the database, the new data is added with a weighted average based on the confidence level C (assuming C=1.0 this time). The new weighted average is (2.598×81.6 + 1.0×70) / (2.598+1) = (212.0 + 70) / 3.598 = 282 / 3.598 ≈78.4.

[0054] Step S5: Signal Switching and Beam Switching Decision This step supports decision logic for two application modes, such as... Figure 4 As shown.

[0055] Mode 1: Repeater signal switch control Within the statistical duration T, the percentage of time P_L during which the real-time S is below the low threshold S_Lth is statistically recorded. If P_L > the set low threshold duration percentage P_Lth, then output the repeater signal start command; Within the statistical duration T, the percentage of time P_H during which the real-time S is higher than the high threshold S_Hth is statistically recorded. If P_H ≥ the set high threshold duration percentage P_Hth, then output a repeater signal shutdown command.

[0056] For example: T = 10 seconds, P_Lth = 60%, low threshold S_Lth = 70; P_Hth = 70%, high threshold S_Hth = 90. If S < 70 for 7 seconds within 10 seconds, the repeater is turned on; if S > 90 for 8 seconds within 10 seconds, the repeater is turned off. This statistical method avoids misjudgments caused by instantaneous fluctuations.

[0057] Mode 2: Phased array antenna beam switching Obtain the real-time comprehensive score S_current for the current beam; Query the historical comprehensive score S_beam_i of other beams at the same location from the fingerprint database; If the beam with the highest score is not the current beam, then output a beam switching command.

[0058] For example: The current beam is Beam_3, S_current=75; a database query reveals that the historical ratings for Beam_7 at this location are 85, Beam_5 is 70, and Beam_3 is 75. Therefore, Beam_7 is selected as the target beam, and the command "Switch to Beam_7" is output.

[0059] Step S6: Data Reporting This step periodically reports the latest smart switch data recorded by the device to the network management system. The reported data includes: Data recording time YYYY-MM-DD HH:MM:SS 2026-03-18 10:23:45 longitude Current longitude coordinates 116123456 latitude Current latitude coordinates 39567890 speed Current speed 350 PCI Current main community 58 RSRP Current reference signal received power -80 SINR Current signal-to-noise ratio 75 Original S Rating before correction 85 Corrected S Revised rating 90 Deviation ∆S Deviation percentage 5.9% Correction time Correction of occurrence time 10:23:46 Signal switch status Current switch status 1 Current beam Current beam 3 Beam to be switched Beam to be switched 5 Credibility Current data credibility 1.0 The reporting cycle is configurable. Local data is recorded every second, and by default, it is reported once every 30 seconds.

[0060] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, for the device embodiments, the above descriptions are merely preferred embodiments of the present invention. Since they are fundamentally similar to the method embodiments, the descriptions are relatively simple, and relevant parts can be referred to the descriptions of the method embodiments. The above descriptions are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention, without departing from the principle of the present invention, should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart scoring and self-learning update method based on a railway vehicle-mounted phased array antenna, characterized in that, Includes the following steps: Step S1: Data Acquisition: Acquire application scenario data, including device location information, reference signal received power (RSRP) of each beam, signal-to-noise ratio (SINR) of each beam, transmit power of each beam, current train speed, downlink channel quality indicator (CQI), and historical handover success rate. Step S2: Initial Comprehensive Score Generation: Based on the application scenario data obtained in Step S1, an initial historical comprehensive score is generated; the formula for calculating the comprehensive score S is: S_Real=W1×F1(RSRP)+W2×F2(SINR)+W3×F3(Distance)+W4×F4(Speed)+W5×F5(Path Loss)+W6×F6(Handover Success Rate)+W7×F7(Dwell Time)+W8×F8(CQI); Wherein, F1(RSRP) is the normalized function of RSRP, F2(SINR) is the normalized function of SINR, F3(distance) is the distance-related function, F4(speed) is the speed-related function, F5(path loss) is the path loss function, F6(handover success rate) is the historical handover success rate function, F7(dwell time) is the beam dwell time function, F8(CQI) is the channel quality indication function, and W1~W8 are weighting coefficients; The dynamic configuration method for the weight coefficients W1 to W8 includes: Scene-based pre-configuration based on line features: different weight combinations are pre-configured for different line scenarios, and automatic switching is achieved through line recognition; Statistical self-learning based on historical data: Statistically analyze the variance distribution of each parameter in historical operation and dynamically adjust the weights; Closed-loop adjustment based on real-time feedback: The actual communication quality after beam switching is used as feedback, and the weights are fine-tuned by gradient descent. A time decay factor λ(t) = e^(-(t-t0) / τ) is introduced to weight the historical data, where τ is a time constant; Step S3: Parameter Configuration: Configure the calculation parameters for the comprehensive score, including the weight of each parameter W1~W8, the intelligent switch threshold S_th, the statistical duration T, the threshold duration percentage P, and the time constant τ; the configuration parameters support local configuration and remote configuration by the network management system. Step S4: Credibility Assessment and Self-Learning Update S4.1 Credibility Assessment: Calculate the comprehensive credibility weight C = C_GPS × C_speed × C_cont for each sampled data, where C_GPS is based on the GPS positioning accuracy factor, C_speed is based on the vehicle speed change rate, and C_cont is based on the sampling time interval; discard the sampled data when C is lower than the preset threshold. S4.2 Self-Learning Update: Using location information as a label, record the historical comprehensive field strength information and beam switching status corresponding to that point; Calculate the closest point from the real-time location to the historical trajectory as a comparison reference point; The historical comprehensive score S_hist of the reference point is calculated by using a time-decay weighted average and weighted by confidence: S_hist=[Σλ(t_i)×C_i×S_i] / Σ(λ(t_i)×C_i); Calculate the percentage deviation ∆S between the real-time scores S_real and S_hist; A graded correction strategy is implemented based on the deviation magnitude and duration. When updating the database, the weighted average is calculated as S = (C_hist × S_hist + C × S_Real) / (C_hist + C). Step S5: Signal switching and beam switching decision: Compare the real-time comprehensive score S with the set threshold S_th: Within the statistical duration T, when the percentage of time below the threshold reaches P, a repeater signal activation command is output. When the percentage of time exceeding the threshold reaches P within the statistical duration T, a repeater signal shutdown command is output. For phased array antenna applications, the real-time S of the current beam is compared with the historical S of other beams, the beam with the highest score is selected as the target beam, and a beam switching command is output. Step S6: Data Reporting: Regularly report the smart switch data to the network management system. The reported data includes: data recording time, location information, field strength information, comprehensive score S automatic correction record, switch switching record, and beam switching status.

2. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that: In the formula for calculating the comprehensive score S in step S2: F1(RSRP)=(RSRP-RSRP_min) / (RSRP_max-RSRP_min), RSRP_min=-120dBm, RSRP_max=-60dBm; F2(SINR)=(SINR-SINR_min) / (SINR_max-SINR_min), SINR_min=0dB, SINR_max=30dB; F3 (distance) = 1 - (d / d_max), where d is the current distance from the base station and d_max is the coverage radius of the base station; F4 (speed) = 1 - (v / v_max), where v is the current speed and v_max = 350 km / h; F5 (path loss) = (PL_max - PL) / (PL_max - PL_min), where PL = transmit power - RSRP, PL_min = 80dB, PL_max = 120dB; F6 (Switch Success Rate) = This location / historical switch success probability of the beam, with a value range of 0 to 1; F7 (Dwell time) = min(Dwell / Dwell_max, 1), Dwell_max = 10 seconds; F8(CQI)=(CQI-CQI_min) / (CQI_max-CQI_min), CQI_min=0, CQI_max=15.

3. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that: The dynamic configuration of weight coefficients in step S2 includes: scenario-based pre-configuration based on route characteristics: different weight combinations are pre-configured for different route scenarios such as tunnels, viaducts, plains, mountains, and urban areas, and the scenarios are automatically identified and weights are switched by matching GPS coordinates with electronic maps; Statistical self-learning based on historical data: Calculate the variance of each parameter, and when the variance of a certain parameter is significantly greater than that of other parameters, reduce the weight of that parameter, and vice versa. Closed-loop adjustment based on real-time feedback: The actual communication quality after beam switching is used as feedback, and the weights are fine-tuned by gradient descent.

4. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that, The specific calculation method for the credibility assessment in step S4.1 is as follows: C_GPS: 1.0 when DOP<2, 0.7 when 2≤DOP<4, 0.4 when DOP≥4, and 0 when there is no GPS signal; C_speed: 1.0 when |Δv / Δt|<2m / s², 0.6 when 2≤|Δv / Δt|<4m / s², and 0.3 when |Δv / Δt|≥4m / s²; C_cont: 1.0 when Δt_samp < 1 second, 0.8 when 1 ≤ Δt_samp < 2 seconds, and 0.5 when Δt_samp ≥ 2 seconds; The overall reliability C = C_GPS × C_speed × C_cont. If C < 0.3, the sampled data is discarded.

5. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that: The hierarchical correction strategy in step S4.2 is as follows: ∆S>30% and short-term (single occurrence): Outlier, no correction, historical S is used; ∆S>30% and lasts for a long time (≥3 consecutive times): permanent environmental change, correct the record, and use real-time S; 10% < ∆S ≤ 30%: Moderate change, correct the record, and use real-time S; 0≤∆S≤10%: Normal fluctuations, no correction is made, real-time S is used.

6. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that: When updating the database in step S4.2, the contribution of the new data is weighted according to the confidence level C, that is, the new weighted average value = (old weighted sum × old total weight + C × S_real) / (old total weight + C).

7. The method according to claim 1, characterized in that, In step S5, the history S of other beams is weighted by time decay and by confidence: S_beam=[Σλ(t_i)×C_i×S_i] / Σ(λ(t_i)×C_i).

8. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that: The statistical self-learning based on historical data in step S2 includes: calculating the variance of each parameter, reducing the weight of a parameter when the variance of a certain parameter is significantly greater than that of other parameters, and increasing the weight of a parameter when the variance of a certain parameter is significantly less than that of other parameters.

9. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that: The default value of the time constant τ in steps S2 and S3 is 7 days, which supports network management configuration.

10. The intelligent scoring and self-learning update method based on a railway vehicle-mounted phased array antenna according to claim 1, characterized in that: The credibility assessment mechanism in step S4.1 includes: calculating GPS credibility based on GPS positioning accuracy factor DOP, calculating speed credibility based on vehicle speed change rate, calculating continuity credibility based on sampling time interval, and the comprehensive credibility weight is the product of the three factors, which are weighted according to credibility when updating the database.

Images