Subway signal fault diagnosis and prediction system based on artificial intelligence

Abstract

The application discloses a subway signal fault diagnosis and prediction system based on artificial intelligence and belongs to the technical field of urban rail transit communication, comprising a data acquisition unit for acquiring the depth of a subway layer, a set of ground weather data above the subway, subway layer feature data and subway passenger flow data, a basic signal analysis unit for generating a subway signal theoretical score, a stay passenger flow analysis unit for generating a stay passenger flow influence factor, a comprehensive analysis unit for generating a subway signal comprehensive score and a diagnosis and prediction unit for carrying out subway signal fault diagnosis and prediction according to the subway signal comprehensive score; the application realizes active prediction of fault risks, improves the accuracy and timeliness of subway signal fault prediction, realizes comprehensive analysis of multidimensional dynamic factors and thus optimizes the deployment efficiency of network resources.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of urban rail transit communication technology, and particularly relates to a subway signal fault diagnosis and prediction system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of urban rail transit, the subway has become the main mode of public travel. The quality of the mobile communication network in the subway station and tunnel (hereinafter referred to as "subway signal") is crucial. When the subway signal fails or the network is congested, it will cause problems such as user call interruption, high network delay, video lag, etc., which seriously affects the user experience.

[0003] In the prior art, the monitoring of the subway signal is mostly based on the performance alarms (such as CPU load, memory usage, error rate, etc.) of the network equipment (such as base stations), which is a passive and lagging monitoring method. Usually, the maintenance personnel only intervene after the fault has occurred and has affected the user. In addition, the quality of the subway signal is affected by complex factors: first, the physical environment, such as the depth of the tunnel, the ground weather (rain, snow, thunderstorm) will attenuate the wireless signal strength; second, user behavior, such as the instantaneous huge flow caused by the morning and evening peak and the hot station transfer will collapse the network capacity.

[0004] Therefore, there is an urgent need for a method that can actively predict signal faults, taking into account environmental and human factors, to provide early warning of subway network congestion and thus provide decision support for network optimization and dynamic resource allocation. SUMMARY

[0005] In view of the deficiencies of the prior art, the application provides a subway signal fault diagnosis and prediction system based on artificial intelligence, which solves the above problems.

[0006] To achieve the above purpose, the application is implemented by the following technical scheme: a subway signal fault diagnosis and prediction system based on artificial intelligence, specifically comprising:

[0007] A data acquisition unit is configured to acquire the depth of the subway layer, the set of ground weather data above the subway, subway layer feature data, and subway passenger flow data, wherein the subway passenger flow data includes the current number of people staying in the subway and the historical passenger flow of the subway.

[0008] A basic signal analysis unit is configured to generate a subway signal theoretical score based on the depth of the subway layer and the set of ground weather data above the subway.

[0009] A stay passenger flow analysis unit is configured to establish a passenger flow analysis model based on the subway layer feature data and the real-time stay passenger flow, and generate a stay passenger flow influence factor.

[0010] The comprehensive analysis unit is configured to generate a subway signal comprehensive score according to the stay passenger flow influence factor and the subway signal theoretical score.

[0011] The diagnostic prediction unit is configured to perform subway signal fault diagnostic prediction according to the subway signal comprehensive score.

[0012] Based on the above technical solutions, the application further provides the following optional technical solutions.

[0013] Further technical solutions: the generation mode of the subway signal theoretical score specifically includes:

[0014] Through the formula:

[0015]

[0016] Generate the subway signal theoretical score ;

[0017] In the formula, represents the normalized value of the i-th weather data in the ground weather data set, represents the weight coefficient of the i-th weather data in the ground weather data set, and m is the number of weather data.

[0018] Further technical solutions: the stay passenger flow analysis unit specifically includes:

[0019] The preliminary prediction module is configured to generate a stay passenger flow preliminary prediction value according to historical subway stay passenger flow.

[0020] The current state analysis module is configured to generate a real-time correction factor according to the current subway stay passenger flow.

[0021] The final stay passenger flow prediction module is configured to generate a final stay passenger flow prediction value according to the stay passenger flow preliminary prediction value, the real-time correction factor, and subway layer feature data.

[0022] The stay passenger flow influence factor generation module is configured to generate a stay passenger flow influence factor according to the final stay passenger flow prediction value by establishing a passenger flow analysis model.

[0023] Further technical solutions: the generation mode of the stay passenger flow preliminary prediction value specifically includes:

[0024] Through the formula:

[0025]

[0026] Generate the stay passenger flow preliminary prediction value ;

[0027] In the formula, represents a smoothing weight coefficient, represents a recent history of the number of people staying, represents a periodic history of the number of people staying.

[0028] Further technical solutions: the generation mode of the real-time correction factor specifically includes:

[0029] Through the formula:

[0030]

[0031] Generate a real-time correction factor ;

[0032] In the formula, represents the current number of people staying in the subway, represents the comfortable capacity threshold of the number of people in the subway station, represents the decay rate correction coefficient.

[0033] Further technical solutions: the generation mode of the final number of people staying prediction value specifically includes:

[0034] Through the formula:

[0035]

[0036] Generate a final number of people staying prediction value ;

[0037] In the formula, represents the preliminary prediction value of the number of people staying, represents the real-time correction factor, represents the subway transfer station sign value, represents the popular station sign value, represents the holiday sign value, , , are weight coefficients, and .

[0038] Further technical solutions: the expression of the people flow analysis model is specifically:

[0039]

[0040] In the expression, represents the final number of people staying prediction value, represents the reference threshold of the network carrying capacity, the unit is the number of people, represents the growth influence coefficient.

[0041] Further technical solutions: the generation mode of the subway signal comprehensive score specifically includes:

[0042] Through the formula:

[0043]

[0044] Generate subway signal comprehensive score ;

[0045] In the formula, Indicates the subway signal theoretical score, Indicates the stay passenger flow influence factor, Indicates the confidence factor.

[0046] Further technical solutions: the generation mode of the confidence factor specifically includes:

[0047] Through the formula:

[0048]

[0049] Generate confidence factor ;

[0050] In the formula, Indicates the normalized value of the historical prediction accuracy, Indicates the sensitivity coefficient.

[0051] Further technical solutions: the subway signal fault diagnosis prediction mode specifically includes:

[0052] The subway signal comprehensive score is compared with the preset score threshold;

[0053] If the subway signal comprehensive score is greater than or equal to the preset score threshold, it is predicted that the subway signal does not have the risk of failure;

[0054] If the subway signal comprehensive score is less than the preset score threshold, it is predicted that the subway signal has the risk of failure.

[0055] The present application provides a kind of subway signal fault diagnosis prediction system based on artificial intelligence, compared with prior art has the following beneficial effects:

[0056] The present application integrates the dynamic interaction of environmental factors and human factors by the synergistic effect of data acquisition unit, basic signal analysis unit, stay passenger flow analysis unit, comprehensive analysis unit and diagnosis prediction unit, generates subway signal comprehensive score using multidimensional data modeling analysis, realizes the active prediction of fault risk, improves the accuracy and timeliness of subway signal fault prediction, realizes multidimensional dynamic factor comprehensive analysis, thereby optimizing the deployment efficiency of network resources. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 A structure schematic diagram of a subway signal fault diagnosis and prediction system based on artificial intelligence is provided for the present application. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0059] The specific implementation of the present application is described in detail below in combination with specific examples.

[0060] Please refer to Figure 1 For an embodiment of the present application, a subway signal fault diagnosis and prediction system based on artificial intelligence is provided, which specifically comprises:

[0061] The data acquisition unit 10 is used to acquire the depth of the subway layer, the ground weather data set above the subway, the subway layer feature data and the subway passenger flow data; wherein the subway passenger flow data includes the current subway remaining number and the historical subway remaining passenger flow;

[0062] The basic signal analysis unit 20 is used to generate a subway signal theoretical score according to the depth of the subway layer and the ground weather data set above the subway;

[0063] The remaining passenger flow analysis unit 30 is used to establish a passenger flow analysis model according to the subway layer feature data and the real-time remaining passenger flow, and generate a remaining passenger flow influence factor;

[0064] The comprehensive analysis unit 40 is used to generate a subway signal comprehensive score according to the remaining passenger flow influence factor and the subway signal theoretical score;

[0065] The diagnosis and prediction unit 50 is used to perform subway signal fault diagnosis and prediction according to the subway signal comprehensive score;

[0066] The depth of the subway layer refers to the vertical distance of the tunnel structure from the ground, which can be realized by a laser range finder or engineering drawing data, and this parameter directly affects the calculation of electromagnetic wave propagation loss;

[0067] The ground weather data set above the subway includes rainfall, lightning intensity and other meteorological indicators, which can be acquired in real time through the meteorological bureau API interface and is used to quantify the interference degree of weather on wireless signal;

[0068] The subway layer feature data includes the station type (whether it is a transfer station) and the popular station type, which can be extracted from the subway operation management system and is used to evaluate the passenger flow of the station;

[0069] The current number of people staying in the subway flow data can be counted by the in-station camera combined with image recognition technology, and the historical staying flow is obtained through the ticket system database, both of which reflect the real-time and periodic flow law.

[0070] Specifically, the data acquisition unit synchronously collects environmental parameters and flow dynamic data, breaking the limitation of a single data source. The basic signal analysis unit establishes a theoretical attenuation model of signal propagation through weighted calculation of depth and weather data, for example, the increase of depth leads to exponential attenuation of signal strength, and heavy rain weather intensifies signal scattering. The staying flow analysis unit combines the site structure characteristics, uses historical data to predict the reference flow, and then corrects the predicted value through real-time staying number, and finally generates a flow influence factor reflecting the network pressure. The comprehensive analysis unit dynamically superimposes the theoretical attenuation and flow load, for example, during the morning and evening peak hours, the weight of the flow influence factor increases, and the comprehensive score significantly decreases. The diagnosis and prediction unit judges the signal state through threshold comparison, and triggers an early warning when the score is lower than the preset threshold, guiding the dynamic expansion of network resources.

[0071] Compared with the prior art, the existing scheme only monitors the performance parameters of the equipment, and cannot predict the influence of environmental and flow changes on signal quality. The present application constructs a two-factor evaluation model of signal quality through the cooperative analysis of environmental parameters and flow data.

[0072] Through the above technical solutions, the present application realizes the dynamic evaluation and fault warning of the subway signal state. In the early morning peak scene of the transfer station, the system can identify the superposition effect of the surge of people flow and signal attenuation in advance, trigger the network resource allocation instruction, and avoid the user's perception of call interruption or video lag. In the emergency situation of tunnel water seepage caused by heavy rain on the ground, the system quickly calculates the signal attenuation rate through weather data and depth parameters, and guides the operation and maintenance personnel to preferentially reinforce the network equipment in the affected area.

[0073] Preferably, the present application further proposes that the generation method of the subway signal theoretical score specifically comprises:

[0074] Through the formula:

[0075]

[0076] generate the subway signal theoretical score ;

[0077] In the formula, is the normalized value of the i-th weather data in the ground weather data set, is the weight coefficient of the i-th weather data in the ground weather data set, and m is the number of weather data;

[0078] wherein the normalized value is referred to as converting weather data of different dimensions into a unified dimensionless value, which can be realized by using the maximum and minimum value normalization method, so that heterogeneous data such as rainfall, ground snow cover rate, lightning intensity, etc. have comparability. Among them, the weight coefficient refers to the contribution degree of different weather types to the signal attenuation of the subway, which can be determined by regression analysis of historical signal attenuation data and weather factors.

[0079] Specifically, after the ground weather data is normalized, the dimensional difference is eliminated, and the multi-source data can be directly weighted and summed. The weight coefficient distribution reflects the difference in the influence of different weather on signal attenuation, for example, the high weight coefficient of thunderstorm weather reflects its significant interference to signal intensity. The summation term represents the total amount of signal attenuation under the superposition of multiple weather factors, and the reciprocal form is used to make the theoretical score value nonlinearly decrease when the adverse weather factor increases, which conforms to the physical law that multiple factors jointly cause the signal intensity to deteriorate sharply in the actual scene. The score result provides an environmental evaluation basis for subsequent comprehensive analysis of passenger flow factors.

[0080] Compared with the prior art, the existing method relies on network device performance alarm for passive monitoring and cannot quantify the influence of weather factors on signal quality. The present application converts weather data into a calculable score index by establishing a mathematical quantitative model, and realizes active prediction of environmental factors. The prior art does not consider the different influence of different weather data on signal attenuation, while the present application accurately reflects the contribution difference of thunderstorm, rainfall and other weather through weight coefficient distribution.

[0081] Through the above technical solutions, the present application solves the problem that the prior art cannot quantify the influence of weather factors on subway signals, realizes the fusion calculation of multi-weather data through normalization processing and weight distribution, generates a theoretical score that can quantify the evaluation of signal intensity, provides an environmental factor analysis basis for subsequent fault prediction, and supports dynamic allocation decision of network resources.

[0082] Preferably, the present application further proposes that the stay passenger flow analysis unit 30 specifically comprises:

[0083] A preliminary prediction module for generating a preliminary stay passenger flow prediction value according to historical subway stay passenger flow;

[0084] A current state analysis module for generating a real-time correction factor according to the current subway stay passenger flow;

[0085] A final stay passenger flow prediction module for generating a final stay passenger flow prediction value according to the preliminary stay passenger flow prediction value, the real-time correction factor and the subway layer feature data;

[0086] The staying passenger flow influence factor generation module is configured to establish a passenger flow analysis model according to the final staying passenger flow prediction value, and generate a staying passenger flow influence factor.

[0087] The historical subway staying passenger flow is a statistical quantity of the number of passengers staying in the subway station in a past time period, and can be stored in a time series database and divided according to a time window, and is used to reflect periodicity and short-term trends of passenger flow.

[0088] The current subway staying passenger flow is real-time collected station passenger data, which can be obtained by using infrared sensors or video recognition technology, and is used to capture sudden changes in passenger flow.

[0089] The subway layer feature data is station attribute data containing transfer station signs, popular station signs and holiday signs, which can be represented by binary coding or classification variables, and is used to distinguish the different effects of different station types on passenger flow distribution.

[0090] The real-time correction factor is an adjustment parameter for nonlinear mapping of the ratio of the current staying passenger flow to the comfort capacity threshold based on a logical function, which can be implemented by using a sigmoid function, and is used to suppress the prediction value deviation when the capacity threshold is exceeded.

[0091] The passenger flow analysis model is a quantitative model for exponential function mapping of the final prediction value and the network carrying reference threshold, which can use the normalized prediction value as an input variable, and is used to represent the dynamic influence of passenger flow overload on signal attenuation.

[0092] Specifically, the preliminary prediction module generates a basic prediction value by fusing recent historical data and periodic historical data, and balancing short-term fluctuations and long-term regularities by using a smoothing weight coefficient. The current state analysis module inputs the real-time staying passenger flow and the preset capacity threshold into the sigmoid function to generate a correction factor reflecting the instantaneous passenger flow pressure. The final prediction module calculates the basic prediction value, the correction factor and the station feature data by weighted superposition, and adjusts the prediction result in combination with the spatial attributes of transfer stations, popular stations and holidays. The staying passenger flow influence factor generation module inputs the final prediction value and the network carrying reference threshold into the exponential function to generate an influence factor that nonlinearly increases with the degree of passenger flow overload, which provides a dynamic correction basis for subsequent signal scoring.

[0093] Compared with the prior art, the existing scheme only relies on passive alarm data of network equipment and cannot predict signal quality decline caused by sudden crowd flow. The application realizes multi-dimensional modeling of crowd flow trend by establishing a dynamic correction mechanism of historical data and real-time data and combining site spatial attribute characteristics. For example, when it is detected that the current remaining number of people exceeds the comfortable capacity threshold, the real-time correction factor automatically reduces the prediction value credibility, avoiding prediction deviation caused by instantaneous crowd surge. At the same time, the exponential function mapping model can accurately reflect the nonlinear influence of crowd overload on signal attenuation, and has higher prediction sensitivity than the traditional linear model.

[0094] Through the above technical scheme, the application solves the problem in the prior art that subway signal quality cannot be actively predicted due to dynamic changes in crowd flow. By fusing historical trend, real-time state and site feature data, a dynamically corrected crowd flow prediction value is generated, and a nonlinear influence factor model is established to accurately quantify the potential influence of crowd distribution on signal quality. For example, during morning and evening peak hours, the system can automatically adjust the prediction value according to the real-time remaining number of people, identify key nodes combined with transfer station signs, and provide data support for signal resource dynamic allocation by early warning possible network congestion.

[0095] Preferably, the application further proposes that the generation mode of the preliminary prediction value of the remaining crowd flow comprises:

[0096] The preliminary prediction value of the remaining crowd flow is generated by the formula:

[0097]

[0098] The preliminary prediction value of the remaining crowd flow is generated by the formula:

[0099] In the formula, represents a smoothing weight coefficient, represents a recent historical remaining crowd flow, represents a periodic historical remaining crowd flow, and periodicity refers to a corresponding time in a time feature cycle, for example, this Friday and last Friday;

[0100] The smoothing weight coefficient is a coefficient for adjusting the contribution proportion of recent data and periodic data in the prediction model, and can be realized by a dynamic adjustment algorithm, for example, different weight values are set based on site type or time period, which balances the influence of sudden events and long-term rules on crowd flow;

[0101] The recent historical remaining crowd flow refers to historical crowd flow data in a recent time period, and can be collected by a sliding window algorithm for data in the last three days or a week, which captures short-term crowd flow fluctuation characteristics;​

[0102] Periodic historical passenger flow refers to historical data with the same periodic attribute as the prediction time point, and specifically, the same day last week or the same day this month can be extracted by using a time series matching algorithm, which reflects the passenger flow law at a fixed time node.

[0103] Specifically, the recent data and periodic data are fused by a linear combination formula, and in the calculation process, the smoothing weight coefficient is dynamically configured to adapt to the prediction needs of different scenarios. For example, in a transfer station, the weight coefficient can be set to a high value to enhance the response ability of the sudden passenger flow change; in a general station, the weight coefficient can be set to a low value to strengthen the stability of the periodic law. Through the adjustable weighting mechanism, the predicted value can reflect both the short-term fluctuations caused by temporary activities or emergencies and the periodic patterns in the historical same period data, thereby avoiding the prediction bias caused by a single data source.

[0104] Compared with the prior art, the traditional method only relies on a single dimension of historical data for prediction, such as only using the data of the recent period or the average value of the fixed period, which cannot cope with the complex spatio-temporal variation characteristics of passenger flow. The present application solves the problem of low prediction accuracy of a single data source by constructing a dual-dimension data fusion model to collaboratively analyze short-term dynamics and long-term regularity, and through the dynamic configurability of the weight coefficient, it realizes the adaptation to different station characteristics and operation scenarios.

[0105] Through the above technical solutions, the present application can effectively predict the dynamic change trend of the passenger flow in the subway station, by fusing the dual characteristics of recent and periodic data, to improve the accuracy and robustness of the prediction results, to provide reliable passenger flow prediction data support for subsequent signal fault diagnosis, to identify the network capacity overload risk in advance and trigger the early warning mechanism.

[0106] Preferably, the present application further proposes that the generation mode of the real-time correction factor specifically comprises:

[0107] Through the formula:

[0108]

[0109] generate a real-time correction factor ;

[0110] In the formula, represents the current subway remaining population, represents the comfortable capacity threshold of the subway station, represents the decay rate correction coefficient;

[0111] wherein the current subway remaining population is the number of people in the station collected in real time by sensors or video recognition technology, which can be realized by infrared counters or smart cameras combined with image analysis algorithms, and is used to reflect the instantaneous state of the current network load;

[0112] the comfortable capacity threshold of the number of people in the subway station is the critical number of people preset according to the space layout of the station and the carrying capacity of the network equipment, which can be determined by multiplying the design capacity of the station by a safety factor, and is used to measure the potential pressure of people flow on network performance;

[0113] the decay rate correction coefficient is a parameter that controls the change rate of the correction factor when the number of people deviates from the threshold, which can be dynamically adjusted by historical data fitting or machine learning model, and is used to adjust the sensitivity of the system to sudden changes in people flow.

[0114] Specifically, the ratio of the current subway remaining number of people to the comfortable capacity threshold is taken as an input variable by a logical function model, and when the actual number of people approaches the threshold, the normalized difference term triggers the nonlinear response of the exponential function, so that the correction factor presents a smooth decay trend with the increase of the number of people. For example, when the actual number of people exceeds the threshold, the positive term in the denominator of the exponential term causes the correction factor to decrease rapidly, thereby inhibiting the excessive growth of the predicted value. This model dynamically associates real-time people flow with the preset capacity, avoiding the prediction mutation problem of traditional linear correction methods in the critical state, and the normalization processing makes the calculation results of different scale stations comparable.

[0115] Compared with the prior art, the traditional method usually adopts a fixed threshold alarm or adjusts the predicted value based on a linear proportion, which cannot accurately reflect the dynamic nonlinear influence of people flow on network capacity. For example, the prior art may only generate a binary alarm signal by simply comparing the current number of people with the threshold, while the continuous correction factor output by the logical function of the present application can not only reflect the deviation degree of people flow, but also quantify its progressive influence on network performance.

[0116] Through the above technical solutions, the present application realizes dynamic quantitative evaluation of real-time people flow in the subway station, solves the problem that the traditional passive monitoring method cannot capture the impact of people flow mutation on network capacity, effectively avoids the violent fluctuation of the predicted value in the critical state, and provides an accurate correction basis for dynamic allocation of network resources. At the same time, the introduction of normalization processing and adjustable decay rate makes the scheme adaptable to the needs of different scale stations and diversified operation scenarios, ultimately improving the timeliness and accuracy of signal fault prediction.

[0117] Preferably, the present application further proposes that the generation mode of the final remaining people flow prediction value comprises:

[0118] Through the formula:

[0119]

[0120] Generating final stay flow prediction value ;

[0121] In the formula, represents a preliminary stay flow prediction value, represents a real-time correction factor, represents a subway transfer station flag value, represents a popular station flag value, represents a holiday flag value, , , are weight coefficients, and ;

[0122] wherein the subway transfer station flag value is an identifier of whether the station is a transfer hub, which can be implemented by a binary variable, for example, a transfer station is marked as 1 and a non-transfer station is marked as 0, for representing the gathering effect of transfer behavior on passenger flow;

[0123] The popular station flag value is an identifier of the commercial activity intensity around the station, which can be implemented by a continuous variable, for example, according to the commercial district passenger flow classification, for reflecting the attraction of commercial activities on passenger flow;

[0124] The holiday flag value is a classification identifier of date attribute, which can be implemented by a classification variable, for example, a statutory holiday is marked as 1 and a weekday is marked as 0, for capturing the peak travel characteristics of holidays;

[0125] The weight coefficient , , is a contribution degree allocation parameter of different scene factors, which can be implemented by a normalization constraint, for example, by constraining the total weight to be 1, to avoid the excessive influence of a single scene factor on the prediction result.

[0126] Specifically, first, a preliminary prediction value is generated based on historical stay flow data, which integrates recent and periodic passenger flow change rules. Then, the current subway stay number is dynamically adjusted by a real-time correction factor, which significantly reduces the prediction value when the real-time number approaches the capacity threshold to reflect the network bearing pressure. Further, three types of scene characteristics, i.e., transfer station, popular station, and holiday, are superimposed, wherein the transfer station flag value strengthens the passenger flow gathering effect of hub stations, the popular station flag value quantifies the attraction of commercial activities to passenger flow, and the holiday flag value identifies the peak travel demand of special dates. Through the weight coefficients 、 、 The influence degree of the three types of scenes is allocated, the contribution proportion of different scenes to the crowd flow prediction is dynamically adjusted under the condition that the weight sum is 1, and finally the prediction values of the comprehensive environment state, real-time capacity and scene characteristics are generated.

[0127] Compared with the prior art, the traditional method only relies on the equipment performance index or static historical data for prediction, and cannot capture dynamic human factors such as instantaneous passenger flow surge at transfer stations and travel mode mutation during holidays. The present application encodes dynamic factors such as transfer behavior, commercial activities and holiday characteristics into quantifiable parameters by introducing a scene marker value and a weight distribution mechanism, so that the prediction model can adapt to the crowd change law under different operation scenes.

[0128] Through the above technical solution, the present application solves the signal fault early warning lag problem caused by the traditional prediction method ignoring dynamic human factors. Through the multi-dimensional fusion of the scene marker value and the weight constraint mechanism, precise modeling of complex scenes such as passenger flow aggregation at transfer hubs, peak congestion in commercial areas and travel surge during holidays is realized, thereby improving the timeliness and accuracy of the crowd flow prediction, and providing reliable data support for capacity warning and resource scheduling of the subway signal network.

[0129] Preferably, the expression of the crowd flow analysis model is further proposed in the present application as follows:

[0130]

[0131] In the expression, represents the final stay crowd flow prediction value, represents the reference threshold of network carrying capacity, and the unit is the number of people, represents the growth influence coefficient;

[0132] The final stay crowd flow prediction value refers to the predicted crowd flow value calculated based on historical data and real-time crowd flow state, which can be generated by a time series prediction model combined with a real-time correction factor, and is used to reflect the actual crowd flow that may occur in the future period;

[0133] The reference threshold of network carrying capacity refers to the maximum crowd flow that can be carried by the network determined according to the base station hardware configuration, channel bandwidth and signal coverage range, which can be set by network stress testing or historical operation data analysis, and is used to represent the upper limit of the physical network capacity of a specific site;

[0134] The growth influence coefficient is a regulation parameter for controlling the relationship between the degree of passenger flow overload and signal quality attenuation, which can be obtained by training historical fault data through a machine learning algorithm, and is used to dynamically adjust the response strength of the model to passenger flow changes.

[0135] Specifically, the deviation of the predicted passenger flow from the network carrying reference threshold is converted into an exponential growth influence factor. When the predicted passenger flow is lower than the reference threshold, the influence factor tends to 1, indicating that the passenger flow load is within the normal network carrying range. When the predicted passenger flow exceeds the reference threshold, the rapid decay characteristic of the exponential term causes the denominator term to decrease sharply, resulting in a sharp increase in the influence factor value towards the value of 2, accurately reflecting the exponential negative impact of passenger flow overload on signal quality. The growth influence coefficient can adapt to the sensitivity differences of different site network devices by adjusting the slope of the exponential term, for example, a larger coefficient is set at a transfer hub station to strengthen the early warning response, and a smaller coefficient is set at an ordinary station to avoid false positives.

[0136] Compared with the prior art, the traditional method uses a fixed threshold alarm mechanism, which triggers an early warning only when the real-time passenger flow exceeds the preset threshold, and cannot quantify the nonlinear relationship between the overload degree and the signal quality attenuation. The existing linear regression model simply relates passenger flow to signal quality, ignoring the marginal effect of decreasing characteristics of network carrying capacity with increasing overload. The present application establishes a dynamic mapping relationship between the passenger flow overload ratio and the signal quality attenuation degree by introducing a logistic function with saturation characteristics, which can not only maintain baseline evaluation when not overloaded, but also accurately reflect the accelerating deterioration trend of network performance in the overload phase.

[0137] Through the above technical solution, the present application realizes dynamic quantitative evaluation of network carrying pressure in the instantaneous passenger flow surge scenario, solves the problem that the traditional method cannot accurately depict the nonlinear relationship between passenger flow and signal quality. Through the combined application of the reference threshold and the growth coefficient, the evaluation sensitivity can be automatically adjusted for different site network conditions, providing differentiated judgment basis for signal fault prediction. Based on the nonlinear conversion mechanism of the exponential function, an early warning signal can be generated when the passenger flow approaches the critical value, which can gain a disposal time window for dynamic allocation of network resources.

[0138] Preferably, the present application further proposes that the generation mode of the subway signal comprehensive score specifically includes:

[0139] Through the formula:

[0140]

[0141] generate a subway signal comprehensive score ;

[0142] In the formula, indicates the subway signal theoretical score, represents a stay flow influence factor, represents a confidence factor;

[0143] wherein the subway signal theoretical score refers to a theoretical signal quality score calculated based on the depth of the subway layer and the ground weather data, which can be realized by using the weighted reciprocal of the normalized weather data and the weight coefficient, and is used to quantify the attenuation effect of environmental factors on signal strength;

[0144] stay flow influence factor refers to a dynamic influence coefficient generated based on a people flow prediction model, which can be calculated by a logical function relationship between the final stay flow prediction value and the network carrying capacity benchmark threshold, and is used to reflect the negative offsetting effect of people flow overload on signal quality;

[0145] confidence factor refers to a correction coefficient generated based on historical prediction accuracy, which can be calculated by a logical function of the normalized historical accuracy and the sensitivity coefficient, and is used to enhance the credibility of the comprehensive score.

[0146] Specifically, by inversely correlating the subway signal theoretical score with the people flow influence factor, the superimposed effects of environmental attenuation and people flow pressure on signal quality are embodied. For example, when the people flow prediction value exceeds the network carrying benchmark threshold, the influence factor increases, resulting in a decrease in the comprehensive score, thereby warning the signal quality risk. At the same time, the confidence factor positively weights the score through the historical prediction accuracy, and when the historical accuracy is high, the credibility of the comprehensive score is improved. The scheme fuses the environmental attenuation theoretical value, real-time people flow pressure and historical prediction reliability to establish a multi-dimensional scoring model, so that the signal quality evaluation contains not only the dynamic changes of the physical environment, but also the user behavior trends, and can dynamically adjust the result weight according to the historical data credibility.

[0147] Compared with the prior art, the existing method only relies on network device performance indicators for passive monitoring, and cannot predict in advance the signal quality deterioration caused by environmental or people flow mutations. The present application actively models the coupling relationship between environmental factors and people flow factors, and combines historical prediction reliability correction to realize dynamic prediction of signal failure risk. For example, in the scenario of heavy rain on the ground superimposed with peak period people flow surge, the existing technology can only trigger an alarm after the base station load is abnormal, while the present application can identify the signal attenuation trend in advance through the comprehensive score, providing a warning window for resource allocation.

[0148] By the technical solution, the application solves the problem of lagging fault prediction caused by single dependence on equipment performance data in the prior art, and realizes dynamic comprehensive evaluation of metro signal quality through joint modeling of environment, passenger flow and historical reliability. For example, in the holiday large passenger flow scenario of a transfer station, the system can timely reflect the network capacity pressure through the passenger flow influence factor, correct the signal attenuation expectation in combination with the theoretical score calculated based on the weather data, and adjust the score reliability according to the historical prediction accuracy, so that the signal fault risk can be identified in advance and the optimization measures can be triggered before the equipment performance is abnormal.

[0149] Preferably, the application further proposes that the generation mode of the confidence factor specifically comprises:

[0150] The confidence factor is generated by the formula:

[0151]

[0152]

[0153] In the formula, represents a normalized value of a historical prediction accuracy, represents a sensitivity coefficient.

[0154] The normalized value of the historical prediction accuracy is a value obtained by standardizing the accuracy after comparing the historical prediction result with the actual fault occurrence, and the original accuracy can be mapped to the interval of 0 to 1 by using the maximum and minimum value normalization method, so as to eliminate the influence of the dimension difference of the accuracy in different time spans on the calculation. The sensitivity coefficient is a parameter for adjusting the change rate of the historical prediction accuracy to the confidence factor, and the sensitivity of the confidence to the accuracy growth can be controlled by adjusting the curvature of the exponential function, for example, when the sensitivity coefficient takes a larger value, a slight improvement of the accuracy can significantly increase the confidence factor.

[0155] Specifically, the technical solution non-linearly correlates the historical prediction accuracy and the confidence factor through a logical function. After the historical prediction accuracy is normalized and input into the formula, the confidence factor grows slowly when the accuracy is low, rises rapidly in the middle interval, and tends to be flat when approaching the upper limit. The sensitivity coefficient is used to control the steepness of the curve, for example, when the system has a higher requirement for prediction stability, the sensitivity coefficient can be increased so that the accuracy change can be more sensitively reflected in the adjustment of the confidence factor. The dynamically changing confidence factor generated in this way can adaptively correct the reliability of the comprehensive score of the metro signal.

[0156] ​​In some embodiments, the calculation of the historical prediction accuracy rate can adopt a sliding window mechanism, for example, using the prediction results of the past 30 days as the statistical period, and updating the accuracy rate data daily. The value of the sensitivity coefficient can be dynamically configured according to the network operation requirements, for example, a higher sensitivity coefficient is used during the morning and evening peak hours to enhance the response speed of the confidence adjustment, and the coefficient value is appropriately reduced during the flat peak hours to reduce fluctuations.

[0157] Compared with the prior art, the existing scheme usually directly uses a fixed threshold or a linear weighting method to evaluate the reliability of the prediction result, which cannot effectively cope with the scoring instability caused by the fluctuation of the historical prediction accuracy rate. However, the present application realizes the nonlinear adaptive adjustment of the confidence factor with the historical accuracy rate by combining the logic function with the sensitivity coefficient, which avoids excessive sensitivity in the low accuracy rate stage and ensures fast convergence in the high accuracy rate stage.

[0158] Through the above technical scheme, the present application effectively solves the problem of unstable confidence of the comprehensive score caused by the fluctuation of the reliability of the historical prediction result in the subway signal fault prediction. The dynamically generated confidence factor can objectively reflect the cumulative effect of the historical prediction accuracy, so that the signal comprehensive score is automatically corrected for reliability bias in the calculation process, thereby improving the reliability of the fault warning result.

[0159] Preferably, the present application further proposes that the subway signal fault diagnosis and prediction method specifically comprises:

[0160] comparing the subway signal comprehensive score with a preset score threshold;

[0161] if the subway signal comprehensive score is greater than or equal to the preset score threshold, it is predicted that the subway signal has no risk of failure;

[0162] if the subway signal comprehensive score is less than the preset score threshold, it is predicted that the subway signal has a risk of failure.

[0163] The present application also proposes a subway signal fault diagnosis and prediction method based on artificial intelligence, which is applied to the above-mentioned subway signal fault diagnosis and prediction system based on artificial intelligence, and specifically comprises:

[0164] obtaining the depth of the subway layer, the ground weather data set above the subway, the subway layer feature data, and the subway passenger flow data; wherein the subway passenger flow data includes the current subway remaining number and the historical subway remaining passenger flow;

[0165] generating a subway signal theoretical score according to the depth of the subway layer and the ground weather data set above the subway;

[0166] establishing a passenger flow analysis model according to the subway layer feature data and the real-time remaining passenger flow to generate a remaining passenger flow influence factor;

[0167] According to the stay flow influence factor and the subway signal theory score, a subway signal comprehensive score is generated;

[0168] According to the subway signal comprehensive score, subway signal fault diagnosis and prediction are performed.

[0169] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based subway signal fault diagnosis and prediction system, characterized in that, The system specifically comprises: a data acquisition unit configured to acquire a depth of a subway layer, a set of ground weather data above the subway, subway layer feature data, and subway passenger flow data, wherein the subway passenger flow data comprises a current subway remaining population and historical subway remaining population flow; a basic signal analysis unit configured to generate a subway signal theoretical score based on the depth of the subway layer and the set of ground weather data above the subway; a remaining population flow analysis unit configured to establish a passenger flow analysis model based on the subway layer feature data and real-time remaining population flow, and generate a remaining population flow influence factor; a comprehensive analysis unit configured to generate a subway signal comprehensive score based on the remaining population flow influence factor and the subway signal theoretical score; a diagnosis and prediction unit configured to perform subway signal fault diagnosis and prediction based on the subway signal comprehensive score; The remaining population flow analysis unit specifically comprises: a preliminary prediction module configured to generate a remaining population flow preliminary prediction value based on historical subway remaining population flow, wherein the remaining population flow preliminary prediction value is generated by a formula: a current state analysis module configured to generate a real-time correction factor based on the current subway remaining population; ； Generating preliminary prediction values of staying people flow ; In the formula, represents a smoothing weight coefficient, represents a recent history of the number of people staying, represents a periodic history of the number of people staying; a final remaining population flow prediction module configured to generate a final remaining population flow prediction value based on the remaining population flow preliminary prediction value, the real-time correction factor, and the subway layer feature data; a remaining population flow influence factor generation module configured to establish a passenger flow analysis model based on the final remaining population flow prediction value, and generate a remaining population flow influence factor. The subway signal theoretical score is generated by a formula: 2.The metro signal fault diagnosis and prediction system based on artificial intelligence according to claim 1, wherein, The real-time correction factor is generated by a formula: The final remaining population flow prediction value is generated by a formula: ； Generating a metro signal theory score ; In the formula, represents the normalized value of the i-th weather data in the ground weather data set, represents the weight coefficient of the i-th weather data in the ground weather data set, and m is the number of weather data. 3.The metro signal fault diagnosis and prediction system based on artificial intelligence according to claim 1, wherein, The passenger flow analysis model is expressed by: The subway signal comprehensive score is generated by a formula: ； Generating real-time correction factors ; In the formula, represents the current number of people remaining in the subway, represents the comfortable capacity threshold of the number of people in the subway station, represents the decay rate correction coefficient. 4.The metro signal fault diagnosis and prediction system based on artificial intelligence according to claim 1, wherein, The confidence factor is generated by a formula: The subway signal fault diagnosis and prediction is performed by comparing the subway signal comprehensive score with a preset score threshold value, predicting that there is no risk of subway signal failure if the subway signal comprehensive score is greater than or equal to the preset score threshold value, and predicting that there is a risk of subway signal failure if the subway signal comprehensive score is less than the preset score threshold value. ； Generating final dwell volume forecast values ; In the formula, represents the preliminary prediction value of the stay passenger flow, represents the real-time correction factor, represents the subway transfer station sign value, represents the popular station sign value, represents the holiday sign value, , , are weight coefficients, and . 5.The metro signal fault diagnosis and prediction system based on artificial intelligence according to claim 1, wherein, ​ ； In the expression, represents the final stay flow prediction value, represents the reference threshold of network carrying capacity, the unit is the number of people, represents the growth influence coefficient. 6.The metro signal fault diagnosis and prediction system based on artificial intelligence according to claim 1, wherein, ​ ​ ； Generating a metro signal composite score ; In the formula, represents the subway signal theoretical score, represents the stay passenger flow influence factor, represents the confidence factor. 7.The metro signal fault diagnosis and prediction system based on artificial intelligence according to claim 6, wherein, ​ ​ ； Generating a confidence factor ; In the formula, represents a normalized value of a historical prediction accuracy, represents a sensitivity coefficient. 8.The metro signal fault diagnosis and prediction system based on artificial intelligence according to claim 1, wherein, ​ ​ ​ ​

Images