Positioning method and device for poor indoor network quality, and electronic equipment

Abstract

The invention discloses a positioning method and device for poor indoor network quality and electronic equipment, and the method comprises the steps: firstly obtaining a network data source, carrying out the first preprocessing of the network data source, obtaining restoration data, carrying out the time granularity unification processing of the restoration data, obtaining unified data, carrying out the multi-source data aggregation of the unified data, obtaining associated data, and carrying out the positioning of the network data source. The method comprises the steps of obtaining associated data, processing the associated data based on a time dimension algorithm to obtain a first causal graph, processing the associated data based on a static dimension algorithm to obtain a second causal graph, finally obtaining a third causal graph according to the first causal graph and the second causal graph, and automatically positioning indoor network poor quality based on the third causal graph. Therefore, waste of human resources is avoided.

Description

Technical Field

[0001] This application relates to network technology, and includes, but is not limited to, a positioning method, apparatus, and electronic device for poor indoor network quality. Background Technology

[0002] With social development and the progress of the times, the Internet has become a necessity in people's lives, whether outdoors or indoors. However, with the widespread use of the Internet, the diagnosis and location of poor indoor network quality has become very important. This is mainly done by collecting basic performance indicators such as signal strength, latency, and packet loss rate reported by terminal devices. However, relying solely on basic performance indicators leads to poor accuracy. How to achieve more accurate location has become an urgent problem to be solved.

[0003] Traditional solutions involve manual setting and step-by-step manual investigation for location, which leads to a waste of human resources. Summary of the Invention

[0004] In view of this, embodiments of this application provide a positioning method, apparatus, and electronic device for poor indoor network quality.

[0005] The technical solution of this application embodiment is implemented as follows: This application provides a method for locating poor indoor network quality. The method includes: acquiring a network data source; performing a first preprocessing on the network data source to obtain repair data; performing time granularity unification processing on the repair data to obtain unified data; performing multi-source data aggregation on the unified data to obtain associated data; processing the associated data based on a time-dimensional algorithm to obtain a first causal graph; processing the associated data based on a static-dimensional algorithm to obtain a second causal graph; obtaining a third causal graph based on the first and second causal graphs; and locating poor indoor network quality based on the third causal graph.

[0006] Optionally, the unified data is aggregated from multiple sources to obtain associated data, including: the unified data includes at least: device ID and timestamp; the associated data is obtained based on the device ID and the timestamp.

[0007] Optionally, the step of processing the associated data based on the time dimension algorithm to obtain the first causal graph includes: processing the associated data based on the No Tears algorithm to obtain the first causal graph.

[0008] Optionally, the step of processing the associated data based on the static dimensionality algorithm to obtain the second causal graph includes: processing the associated data based on the TTPM algorithm to obtain the second causal graph.

[0009] Optionally, locating poor indoor network quality based on the third causal graph includes: processing the third causal graph based on a first preset strategy to obtain a final anomaly score and topology impact coefficient.

[0010] Optionally, after obtaining the final anomaly score and topological influence coefficient, the process includes: obtaining edge weights based on the final anomaly score and the topological influence coefficient; and processing the edge weights based on a preset random walk strategy to obtain the root cause probability.

[0011] Optionally, the edge weights are processed based on a preset random walk strategy, including: the preset random walk strategy includes at least: a first-order random walk and a second-order random walk; the first-order random walk generates the next hop probability based on the edge weights; the second-order random walk introduces a historical node memory mechanism and balances the transfer weights of the current and previous nodes through a first preset parameter.

[0012] Optionally, after processing the edge weights based on a preset random walk strategy, the process includes: obtaining a directed graph based on the root cause probability and the final anomaly score; and processing the directed graph based on the Dijkstra inverse search algorithm to obtain the localization result.

[0013] A positioning device, comprising: an acquisition unit, an analysis unit, and a processing unit; the acquisition unit being used to acquire a network data source; the analysis unit being used to perform a first preprocessing on the network data source to obtain repair data; perform time granularity unification processing on the repair data to obtain unified data; perform multi-source data aggregation on the unified data to obtain associated data; process the associated data based on a time-dimensional algorithm to obtain a first causal graph; process the associated data based on a static-dimensional algorithm to obtain a second causal graph; the processing unit being used to obtain a third causal graph based on the first and second causal graphs, and to locate poor indoor network quality based on the third causal graph.

[0014] An electronic device includes: a memory for storing at least one set of instructions; a processor for acquiring a network data source; performing a first preprocessing on the network data source to obtain repair data; performing time-granularity unification processing on the repair data to obtain unified data; performing multi-source data aggregation on the unified data to obtain correlated data; processing the correlated data based on a time-dimensional algorithm to obtain a first causal graph; processing the correlated data based on a static-dimensional algorithm to obtain a second causal graph; obtaining a third causal graph based on the first and second causal graphs; and locating poor indoor network quality based on the third causal graph.

[0015] This invention provides a method, apparatus, and electronic device for locating poor indoor network quality. First, a network data source is acquired. Then, the network data source undergoes a first preprocessing step to obtain repair data. Next, the repair data undergoes time-granularity unification processing to obtain unified data. Then, the unified data undergoes multi-source data aggregation to obtain correlated data. The correlated data is processed based on a time-dimensional algorithm to obtain a first causal graph. The correlated data is then processed based on a static-dimensional algorithm to obtain a second causal graph. Finally, based on the first and second causal graphs, a third causal graph is obtained. The method automatically locates poor indoor network quality based on the third causal graph, thereby avoiding waste of human resources. Attached Figure Description

[0016] Figure 1 A flowchart of a method for locating poor indoor network quality provided in an embodiment of the present invention; Figure 2 Another flowchart of the indoor network quality poor location method provided in an embodiment of the present invention; Figure 3 Another flowchart of the indoor network quality poor location method provided in an embodiment of the present invention; Figure 4 Another flowchart of the indoor network quality poor location method provided in an embodiment of the present invention; Figure 5 Another flowchart of the indoor network quality poor location method provided in an embodiment of the present invention; Figure 6 Another flowchart of the indoor network quality poor location method provided in an embodiment of the present invention; Figure 7 Another flowchart of the indoor network quality poor location method provided in an embodiment of the present invention; Figure 8 A schematic diagram of the positioning device provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structural composition of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] 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.

[0018] Please refer to Figure 1 ,in, Figure 1 A flowchart illustrating an implementation of the indoor network quality poor location method provided in this application embodiment may include: Step S101: Obtain the network data source; Step S102: Perform first preprocessing on the network data source to obtain repaired data; Step S103: Perform time granularity unification processing on the repaired data to obtain unified data; Step S104: Aggregate the unified data from multiple sources to obtain related data; Step S105: Process the associated data based on the time dimension algorithm to obtain the first causal graph; Step S106: Process the associated data based on the static dimensionality algorithm to obtain the second causal graph; Step S107: Obtain the third causal graph based on the first and second causal graphs, and locate the poor indoor network quality based on the third causal graph.

[0019] Among them, the network data sources include at least: gateway performance data, home networking performance data, home network set-top box performance data, etc.; the first preprocessing includes at least: data cleaning and outlier detection and processing; time granularity unification is multi-time granularity data unification; the associated data is multi-source data aggregation; the first causal graph is causal graph B, the second causal graph is causal graph A, and the third causal graph is causal graph C.

[0020] Specifically, 1. Selection of data source for home indoor network quality diagnosis.

[0021] 1.1 Data selection for home indoor business scenarios; To comprehensively diagnose home network services, this solution covers three key service scenarios: gateway services, home networking services, and home network set-top box services. A comprehensive assessment of home network quality is achieved through joint modeling of gateway performance data, home networking performance data, and home network set-top box performance data. For gateway services, the focus is on key performance indicators such as gateway device performance information (CPU utilization, memory utilization) and device traffic metrics (downlink traffic peak per cycle, downlink bandwidth peak per cycle). For home networking services, the focus is on analyzing parameters such as WLAN signal strength and network device performance information (same as gateway). For home network set-top box services, the focus is on user experience-related data such as the total duration of buffering within a period and the total duration of screen flickering within a reporting period. By integrating and analyzing this data, the root causes of poor network quality can be more accurately located, providing strong support for subsequent optimization and improvement.

[0022] For the data from the three business scenarios, this solution primarily selected the following key business performance data as the main sources of algorithm training data: (a) Gateway service data: CPU utilization: This metric reflects the CPU usage of the gateway device. High utilization may lead to decreased device performance and affect network connection quality.

[0023] Memory usage: This metric reflects the memory usage of the gateway device. High memory usage may cause slow device response, which in turn affects the stability of network services.

[0024] Service bandwidth: This metric represents the amount of service data traffic processed by the gateway and is crucial for assessing network congestion and bandwidth utilization. Excessive service bandwidth can lead to increased network latency and negatively impact user experience.

[0025] Transmitted optical power: Transmitted optical power refers to the intensity of the optical signal emitted by an optical module or fiber optic device when transmitting data, usually measured in dBm (decibel milliwatt). This indicator directly affects the transmission quality and distance of the optical signal and is one of the key parameters in optical communication systems.

[0026] Received optical power: Received optical power refers to the intensity of the optical signal received by the optical module or fiber optic equipment (unit: dBm). Too high a value may overload the optical module, while too low a value may lead to signal loss or an increase in the bit error rate. It must be kept within the range specified by the equipment (e.g., -23dBm to -8dBm).

[0027] Gateway uplink bandwidth periodic average: This metric represents the gateway's average uplink bandwidth usage over a statistical period, reflecting the average load of user data uploads. Consistently approaching the bandwidth limit may lead to network congestion.

[0028] Gateway downlink bandwidth average over a period of time: This metric represents the average downlink bandwidth usage of the gateway within a statistical period, reflecting the average load of user data downloads. If it remains close to the bandwidth limit for an extended period, it may lead to increased network latency.

[0029] Gateway uplink traffic peak during a given period: This metric represents the highest value of gateway uplink traffic within a given statistical period. It is used to identify sudden traffic spikes or abnormal upload behavior (such as DDoS attacks or large-scale data synchronization).

[0030] Gateway downlink traffic peak during a given period: This metric represents the highest value of downlink traffic to the gateway within a given statistical period. It is used to detect sudden download events (such as video buffering or large file downloads) or abnormal traffic.

[0031] Channel noise floor: The level of background noise in the wireless channel (unit: dBm). A high noise floor will reduce the signal-to-noise ratio (SNR), leading to a decrease in wireless speed or unstable connection. It is usually required to be below -90dBm to ensure normal communication.

[0032] Number of failed transmissions: This counts the number of data packets that the gateway failed to send during data transmission. Frequent failures may be caused by network congestion, signal interference, or equipment malfunction, and should be analyzed in conjunction with packet loss rate.

[0033] Number of received error packets: The number of data packets that failed verification or were corrupted received by the gateway. A high number of error packets may be caused by line interference, abnormal optical power, or equipment compatibility issues.

[0034] Average uplink traffic per cycle for connected devices: The average uplink traffic of terminal devices (such as mobile phones and computers) connected to the gateway within a cycle.

[0035] Average downlink traffic of connected devices over a period of time: This counts the average downlink traffic of terminal devices connected to the gateway within a period, which can help identify devices with high bandwidth usage or abnormal download behavior.

[0036] Peak uplink traffic of downstream devices during a given period: The peak uplink traffic of a single downstream device within a statistical period, used to locate the source of sudden traffic spikes.

[0037] Downlink traffic peak of a single device during the statistical period: The peak downlink traffic of a single device within the statistical period can reveal large-volume download behavior.

[0038] WLAN signal strength: The Wi-Fi signal strength received by the terminal device (unit: dBm). Generally, -30dBm to -60dBm is considered a good signal, while below -70dBm may lead to unstable connections. AP deployment needs to be optimized based on channel interference conditions.

[0039] Number of disconnections: The number of abnormal disconnections made by the gateway or connected devices within the statistical period. Frequent disconnections may be caused by weak signal, interference, IP conflicts, or equipment failure.

[0040] Bit error rate of LAN port received traffic: The percentage of bit errors when data is received via the wired LAN interface. A high bit error rate (e.g., >0.1%) may be caused by aging network cables, port failures, or electromagnetic interference, and the physical layer connection needs to be checked.

[0041] Error frames in LAN port received traffic: The number of frames received by the LAN interface that are malformed or have failed CRC checks. A persistent stream of error frames may indicate a problem with the switch, network interface card (NIC), or cable.

[0042] DNS response latency: The average time (in milliseconds) for a gateway to perform DNS domain name resolution. High latency (e.g., >100ms) may cause slow webpage loading, requiring checking of DNS server configuration or network link quality.

[0043] TCP retransmission rate: The proportion of data retransmissions triggered by the TCP protocol at the transport layer due to packet loss or timeout. A high retransmission rate (e.g., >1%) indicates network congestion or unstable links, requiring route optimization or investigation of the cause of packet loss.

[0044] TTL information: The Time-To-Live (TTL) value in the data packet, representing the maximum number of router hops traversed.

[0045] (b) Network set-top box service data: CPU utilization: Percentage of CPU resources used by the set-top box. Sustained high utilization may cause lag and decoding delays; background processes or video decoding load should be checked.

[0046] Memory usage: Percentage of set-top box memory used. Insufficient memory will trigger frequent garbage collection, causing application crashes or slow interface response. Application memory management needs to be optimized.

[0047] Wireless signal strength: Wi-Fi signal strength connected to the set-top box (unit: dBm).

[0048] Total stuttering duration within this statistical period: The cumulative duration of video playback stuttering within the period (unit: seconds). Stuttering is usually caused by network jitter (such as UDP packet loss), insufficient bandwidth (below the video bitrate requirement), or insufficient terminal decoding performance.

[0049] Total number of program plays during this statistical period: The total number of video-on-demand / live streaming requests initiated by users during the period, used to assess business activity. Abnormal surges may be caused by automated streaming or abnormal requests.

[0050] Total number of successful program playbacks within this statistical period: Number of programs successfully played within the period (HTTP 200 / 206 response).

[0051] Average latency for m3u8 file request and response: The average download time (in milliseconds) for m3u8 index files in the HLS protocol.

[0052] Average latency for media file request response: The average download time (in milliseconds) for video segments (such as TS / MP4 clips).

[0053] EPG Request Response Average Latency: The average response time (in milliseconds) of the Electronic Program Guide (EPG) data interface.

[0054] Average TCP connection establishment time: The average time (in milliseconds) for the set-top box to establish a TCP connection with the server.

[0055] Average retransmission rate: The proportion of TCP packets retransmitted at the transport layer (number of retransmitted packets / total number of packets).

[0056] Total duration of screen flickering within this reporting period: The cumulative duration (in seconds) of mosaic / color blocks appearing in the video footage within the period.

[0057] The maximum value of multiple first-load durations within this period: the maximum value of the first-load duration (from click to first frame display) of all playback sessions within the period (unit: ms).

[0058] (c) Home networking service performance data: This solution involves network devices including wireless routers and access points, which together determine the coverage and signal strength of the home network. The solution will utilize all data information regarding uplink status, downlink devices, Wi-Fi status, and the surrounding Wi-Fi environment.

[0059] 1.2 Selection of data collection period; When selecting data collection periods, it's essential to consider both the actual usage of the home network and diagnostic needs. First, data collection should be conducted during peak network usage times, such as 6 PM to 10 PM, when most family members are home and use the network frequently, providing a more comprehensive picture of poor network quality. Second, considering that family members' activities may differ on weekends and holidays, network data should also be collected during these times to obtain more complete network status information.

[0060] The aforementioned time-period strategy will be directly applied to the core data sources upon which subsequent positive and negative sample construction depends—the gateway, network, and set-top box performance data corresponding to SA board quality reports, user complaints, and low satisfaction data—to ensure the capture of real quality-poor scenarios.

[0061] 1.3 Selection of positive and negative sample sources; Based on the network management, networking, and set-top box data collected during the key time periods selected in step (2), this scheme selects the following three categories as sources of users with poor network quality (negative samples): (a) Reporting of poor quality of SA boards (equipment-side data) Source data: Poor quality indicators of gateways and networking equipment reported by SA boards during peak periods (such as high bit error rate and weak signal).

[0062] Reasons: Highly objective, it can accurately locate problems such as poor signal coverage and hardware failure.

[0063] (b) User complaints (proactive feedback data) Source data: Network management, networking, and set-top box data associated with the complaint period (such as set-top box lag indicators when users reported "video buffering at night").

[0064] Reason: High-priority issue, which can help locate peak load or interference.

[0065] (c) Low satisfaction data (experience-side data) Source data: Device performance data of users with low ratings during the corresponding time period (such as correlation analysis between gateway CPU peak and satisfaction).

[0066] Reason: Quantify subjective experience and identify soft issues (such as insufficient bandwidth) that SA did not capture.

[0067] By integrating data collected during key periods from the device side (SA), feedback side (complaints), and experience side (satisfaction), the entire "network-device-user" chain can be fully covered, accurately pinpointing the root causes of poor quality.

[0068] For positive samples, this scheme adopts the following screening criteria: user data with no poor network quality records for three consecutive months, specifically including: No abnormalities were found on the equipment side: the SA board did not report any poor quality indicators during the monitoring period (such as bit error rate and signal strength remained within the normal range). Key performance indicators of gateways and networking equipment (such as latency and packet loss rate) continued to meet the standards.

[0069] No user feedback: The target users did not file any network-related complaints within three months. There are no records of low satisfaction ratings (such as network issues not being marked in user surveys or app ratings).

[0070] 2. Processing and aggregation of performance data.

[0071] This section proposes a unified data processing and aggregation method for gateway performance data, set-top box performance data, and smart networking gateway performance data. The specific steps are as follows: 2.1 Data cleaning and outlier handling; (1) Data cleaning: The system corrects missing values, duplicate values, and format errors in the original data. Missing values ​​are filled using linear interpolation between preceding and following time points; duplicate data is deduplicated by timestamp; and non-numerical data (such as device status text) is converted to standardized encoding.

[0072] For multi-source heterogeneous data (such as differences in metric naming among different manufacturers), establish unified field mapping rules, for example, unify "CPU_Usage" and "CPU utilization" as "cpu_utilization".

[0073] (2) Outlier Detection and Handling: Outliers are identified using statistical methods (such as the 3σ principle or interquartile range method) and verified using business logic (e.g., CPU utilization exceeding 100% is considered invalid).

[0074] For confirmed outliers, threshold truncation (e.g., replacing values ​​exceeding the 99th percentile with the 99th percentile value) or removal followed by interpolation repair is used to ensure data rationality.

[0075] 2.2 Unification of multi-time granularity data; To address the potential issue of multiple time granularities (minutes, hours, days, etc.) in the original data, the following steps should be taken to unify them: (1) Granularity alignment: Based on the smallest time granularity (such as minute level), generate minute-level virtual data by linear decomposition of coarse-grained data (such as hour level) to ensure time axis alignment.

[0076] (2) Indicator Aggregation: For indicators that need to be aggregated (such as traffic and load), calculate the average (reflecting the overall trend) or the maximum (reflecting peak pressure) at the target granularity (such as hourly). For example: Hourly average: take the arithmetic mean of 60 minutes of data; Hourly maximum: take the peak value within 60 minutes.

[0077] 2.3 Multi-source data aggregation; After cleaning and handling outliers of gateway, set-top box, and smart networking data, as well as unifying data across multiple time granularities, the next step is to aggregate multi-source data. This step is crucial, as it integrates data scattered across different data sources, in different formats, and reflecting the performance of different devices, forming a comprehensive, unified, and correlated performance data table. This provides a solid data foundation for subsequent data analysis, model training, and network quality diagnosis.

[0078] Based on the aggregation of multi-source data, related data is obtained; then, the related data is processed based on the time dimension algorithm to obtain the first causal graph; the related data is processed based on the static dimension algorithm to obtain the second causal graph; based on the first and second causal graphs, the third causal graph is obtained; and the poor indoor network quality is located based on the third causal graph.

[0079] Please refer to Figure 2 The method in this embodiment may include: aggregating unified data from multiple sources to obtain related data, including: Step S201: The unified data shall include at least: device ID and timestamp; Step S202: Obtain associated data based on device ID and timestamp.

[0080] Specifically, during data aggregation, the association between data is primarily established based on two key pieces of information: device ID and timestamp. The device ID is a unique identifier for each device (such as a gateway, set-top box, or networking device), clearly distinguishing different individual devices and ensuring that various types of data from the same device can be accurately associated. The timestamp, on the other hand, records the specific moment the data was generated, providing consistency and synchronization over time.

[0081] The specific operation process is as follows: First, regarding gateway performance data, which includes numerous indicators reflecting the operating status of the gateway device, such as CPU utilization, memory utilization, service bandwidth, transmit optical power, receive optical power, average uplink bandwidth per cycle, and peak downlink bandwidth per cycle, etc. These data all carry their corresponding device IDs and precise timestamp information. For example, a gateway device with ID "GW001" recorded its CPU utilization as 30% and memory utilization as 40% at 10:00:00 on October 1, 2024.

[0082] Set-top box performance data also includes device IDs and timestamps. The metrics covered include CPU utilization, memory usage, Wi-Fi signal strength, total stuttering time within the current statistical period, and average latency for m3u8 file requests and responses. For example, assuming a set-top box with device ID "STB001" records a CPU utilization of 25%, a memory usage of 35%, and a total stuttering time of 5 seconds at 10:05:00 on October 1, 2024. Smart network gateway performance data is similar, including uplink status, downlink devices, Wi-Fi status, and surrounding Wi-Fi environment information, all bound to device IDs and timestamps. For instance, a smart network device with device ID "WN001" records normal uplink status, 5 downlink devices connected, and a Wi-Fi signal strength of -50dBm at 10:02:00 on October 1, 2024.

[0083] During the aggregation process, the system traverses all cleaned data from gateways, set-top boxes, and smart network devices. When data with the same device ID and matching timestamps is encountered, performance data from different data sources corresponding to the same device at the same time is integrated together. For example, if data for gateway device "GW001" at 10:00:00 on October 1, 2024, is found, along with data for set-top box "STB001" and smart network device "WN001" in the same home network matching that time, the system will merge them into a single record. This record will contain all performance metrics of the gateway at that time, the corresponding performance metrics of the set-top box, and the relevant performance metrics of the smart network device, thus forming a complete dataset reflecting the overall performance of the home network at that specific moment. In this way, all data is continuously correlated and integrated, ultimately generating a unified performance data table. This data table comprehensively covers the performance of various key devices in the home network at different points in time, providing rich and effective data support for subsequent in-depth analysis of home network quality and accurate identification of the root causes of network problems.

[0084] Please refer to Figure 3 The method in this embodiment may include: Step S301: Process the associated data based on the No Tears algorithm to obtain the first causal graph; Step S302: Process the associated data based on the TTPM algorithm to obtain the second causal graph; Step S303: Obtain a third causal graph based on the first and second causal graphs, and locate poor indoor network quality based on the third causal graph.

[0085] Specifically, 3. Learning the causal relationships between features.

[0086] 3.1 Learning the causal structure of non-time-series data; The data output in Section 2.3 is used as the training set for causal learning. The timestamps in the dataset are removed, and each data point is treated as an independent data point. The No Tears algorithm is applied to construct the causal structure.

[0087] No Tears algorithm: It transforms causal graph learning into a continuous optimization problem. It directly learns the directed acyclic graph (DAG) through algebraic constraints (such as the DAG property of the adjacency matrix) and generates the causal graph A.

[0088] Output: A causal graph A between features, where each node in the causal graph represents all the features mentioned in the table in Section 1.

[0089] 3.2 Causal Structure Learning of Time Series Data; The data output in Section 2.3 is used as the training set for causal learning. The timestamp data is retained, and the data is now time series data. The TTPM algorithm is used for modeling.

[0090] Topological Hawkes Processes (TTPM) is a causal reasoning algorithm that combines topological structures and Hawkes processes, primarily used for mining causal relationships in event sequence data. Its core principles can be summarized in the following two aspects: I. Basic Model: Hawkes Process TTPM is based on the Hawkes process, which quantifies the impact of historical events on the probability of triggering the current event through a decaying kernel function (such as an exponential function). For example, after event A occurs, its impact decays over time, but it may still increase the probability of subsequent events B occurring.

[0091] II. Topological Causality Discovery TTPM's core innovation lies in the introduction of topological constraints: Structural modeling: Associating event nodes according to the topological network, and representing potential causal relationships through edge connections.

[0092] Dynamic learning: The EM algorithm is used to iteratively adjust the causal graph structure (adding / deleting / reversing edges) to maximize the likelihood probability of the observed data.

[0093] Parameter optimization: Simultaneously adjust parameters such as the attenuation coefficient and BIC penalty factor to balance model complexity and fit.

[0094] Output: A causal graph B between features, where each node in the causal graph represents all the features mentioned in the table in Section 1.

[0095] 3.3 Fusion of the outputs of the cause-effect graph algorithm; Based on the following indicators, cause-effect graphs A and B are comprehensively evaluated, the optimal result is selected, and a preliminary cause-effect graph C is output.

[0096] Structural rationality: Check the properties of the DAG (whether there is a directed cycle), the degree of fit between the number of edges and the actual business logic (such as whether the causal chain of "hardware configuration → CPU average load" conforms to the operating principle of the device).

[0097] Computational efficiency: Compare the number of iterations and convergence time of the two models, and give priority to the model with higher stability under large-scale data.

[0098] Then, based on the causal graph C, which is the third causal graph, the poor quality of the indoor network is located.

[0099] 4. Injection of prior knowledge.

[0100] Based on the initial cause-effect diagram C, it was revised and optimized using the experience of business experts to ensure that the cause-effect structure conforms to actual business logic. The specific methods are as follows: 4.1 Business rule constraints; Conflict resolution: When the algorithm results contradict business knowledge (such as "memory usage → CPU temperature" not matching the hardware heat dissipation design), adjust the causal direction based on expert experience.

[0101] Forced causal constraints: Apply hard constraints to known deterministic causal relationships (such as "firmware version → device stability") to ensure that they exist in the final causal graph.

[0102] 4.2 Supplementing latent variables; Modeling of unobserved variables: For causal paths that are not identified by the algorithm but are important to the business (such as "network congestion → video stuttering" may be affected by hidden "router load"), the hidden variable nodes are manually added.

[0103] 4.3 Causal weight calibration; Expert scoring adjustment: Assign weights to causal edges based on business importance (e.g., "insufficient bandwidth → video buffering" has a greater business impact than "disk I / O → startup delay").

[0104] Bayesian probability correction: If there is uncertainty in the causal strength, Bayesian posterior estimation can be performed by combining expert priors.

[0105] The final output cause-effect graph is shown in the example below: { "nodes": [ {"id": "n1", "name": "cpu utilization"}, {"id": "n2", "name": "Memory usage"}, {"id": "n3", "name": "Service bandwidth size"}, {"id": "n4", "name": "Emitted optical power"}, {"id": "n5", "name": "Received optical power"}, {"id": "n6", "name": "Gateway uplink bandwidth periodic average"}, {"id": "n7", "name": "Gateway downlink bandwidth per cycle average"}, {"id": "n8", "name": "Gateway uplink traffic peak period"}, {"id": "n9", "name": "Peak periodic gateway downlink traffic"}, {"id": "n10", "name": "Channel noise floor"}, {"id": "n11", "name": "Number of failed attempts"}, {"id": "n12", "name": "Number of received error packets"}, {"id": "n13", "name": "Average uplink traffic per cycle for downstream devices"} {"id": "n14", "name": "Average downlink traffic per cycle for connected devices"} {"id": "n15", "name": "Peak uplink traffic period of the downstream device"}, {"id": "n16", "name": "Peak downlink traffic period of connected device"}, {"id": "n17", "name": "WLAN signal strength"}, {"id": "n18", "name": "Number of disconnections"}, {"id": "n19", "name": "Bit error rate of LAN port received traffic"}, {"id": "n20", "name": "Number of erroneous frames in the LAN port received traffic"}, {"id": "n21", "name": "DNS response latency"}, {"id": "n22", "name": "TCP retransmission rate"}, {"id": "n23", "name": "CPU utilization"}, {"id": "n24", "name": "Memory usage"}, {"id": "n25", "name": "Wireless signal strength"}, {"id": "n26", "name": "Total duration of buffering during this statistical period"}, {"id": "n27", "name": "Total number of times the program was played during this statistical period"}, {"id": "n28", "name": "Total number of successful program playbacks during this statistical period"} {"id": "n29", "name": "Average latency for m3u8 file request and response"}, {"id": "n30", "name": "Average latency for media file request response"}, {"id": "n31", "name": "EPG request response average latency"}, {"id": "n32", "name": "Average TCP connection establishment time"}, {"id": "n33", "name": "Average retransmission rate"}, {"id": "n34", "name": "Total duration of screen flickering during this reporting period"}, {"id": "n35", "name": "The maximum value of the duration of multiple first loads within this cycle"} ], "edges": [ {"source": "n3", "target": "n1"}, {"source": "n3", "target": "n2"}, {"source": "n13", "target": "n6"}, {"source": "n14", "target": "n7"}, {"source": "n15", "target": "n8"}, {"source": "n16", "target": "n9"}, {"source": "n4", "target": "n12"}, {"source": "n5", "target": "n12"}, {"source": "n10", "target": "n17"}, {"source": "n10", "target": "n11"}, {"source": "n10", "target": "n12"}, {"source": "n19", "target": "n12"}, {"source": "n20", "target": "n12"}, {"source": "n12", "target": "n22"}, {"source": "n11", "target": "n22"}, {"source": "n6", "target": "n22"}, {"source": "n7", "target": "n22"}, {"source": "n17", "target": "n25"}, {"source": "n22", "target": "n33"}, {"source": "n21", "target": "n31"}, {"source": "n8", "target": "n22"}, {"source": "n9", "target": "n22"}, {"source": "n17", "target": "n18"}, {"source": "n22", "target": "n18"}, {"source": "n25", "target": "n32"}, {"source": "n25", "target": "n33"}, {"source": "n33", "target": "n29"}, {"source": "n33", "target": "n30"}, {"source": "n33", "target": "n31"}, {"source": "n32", "target": "n29"}, {"source": "n32", "target": "n30"}, {"source": "n32", "target": "n31"}, {"source": "n29", "target": "n26"}, {"source": "n29", "target": "n35"}, {"source": "n30", "target": "n26"}, {"source": "n30", "target": "n34"}, {"source": "n30", "target": "n35"}, {"source": "n31", "target": "n35"}, {"source": "n23", "target": "n26"}, {"source": "n23", "target": "n34"}, {"source": "n24", "target": "n26"}, {"source": "n24", "target": "n34"}, {"source": "n33", "target": "n34"}, {"source": "n7", "target": "n30"}, {"source": "n18", "target": "n28"}, {"source": "n22", "target": "n30"}, {"source": "n12", "target": "n30"} ] }

[0106] The method in this embodiment may include: locating poor indoor network quality based on a third cause-effect graph, including: processing the third cause-effect graph based on a first preset strategy to obtain a final anomaly score and topology influence coefficient.

[0107] Specifically, 5. Definition of outlier scores (points to be protected).

[0108] After establishing causal graph relationships between nodes, this proposal uses a random walk method to locate the root cause of poor network quality data. In the random walk-based root cause localization system, the final anomaly score A(v,t) is a comprehensive value used to quantify the degree of anomaly of node v at time t. It not only reflects the deviation of the node's own indicators from the normal baseline value, but also incorporates the effects of time persistence, topological location, and device status information. A(v,t) is a key indicator for the system to judge the degree of node anomaly, directly determining the access probability of the node during the random walk and the path weight allocation in reverse path reasoning.

[0109] The basis for calculating the final anomaly score: single-indicator deviation Obtaining anomaly scores for a single indicator typically involves the following steps: Establish a baseline: Calculate the mean μ and standard deviation σ of normal data (historical data or training data).

[0110] Calculate deviation: For the current outlier data point x, calculate the difference between it and the baseline mean.

[0111] Standardization: Divide the difference by the standard deviation to obtain the standardized single-indicator outlier score.

[0112] Relevant formulas:

[0113] Spatiotemporal-topology coupled anomaly scoring model: To address the limitations of single-dimensional anomaly scores in complex networks, this proposal suggests an anomaly scoring model that integrates spatiotemporal and topological characteristics. The model is then dynamically optimized through reinforcement learning to obtain the final anomaly score A(v,t).

[0114] 5.1 Mathematical Model for Outlier Scores Multidimensional indicator fusion (spatial dimension): Suppose there are n key performance indicators, define the i-th indicator as follows:

[0115] in , The dynamic baseline (calculated based on sliding window statistics, using an exponentially weighted moving average and incorporating a forgetting factor) is used. (The calculation method is similar). Through a nonlinear compression function:

[0116] The final spatial dimension score is:

[0117] The time-series decay factor (time dimension) is defined as the duration of the anomaly being t (minutes), and uses a piecewise decay function:

[0118] in λ represents the critical duration (e.g., 5 minutes) and λ is the decay rate parameter.

[0119] 5.2 Topological Influence Coefficient Consider the topology characteristics of node v: d: number of hops from node to core device; Device type weight (e.g., core switch = 2.0, AP = 1.5)

[0120] 5.3 Device Status Gain Assuming the device has m abnormal states, define the state gain factor: ,in For state The gain coefficient (e.g., θ=0.3 in the overheated state).

[0121] 5.4 Final Anomaly Score

[0122] 5.5 Reinforcement Learning Optimization Framework (I) State Space Design

[0123] N: Number of nodes E: Topological edge feature dimension M: Historical record feature dimension (II) Definition of Action Space The agent dynamically adjusts the key model parameters used to calculate the final anomaly score A(v,t):

[0124] Constraints:

[0125] in: Adjusting the weight of indicator i in spatial dimension fusion ,Influence .

[0126] Δλ: Adjusts the decay rate parameter λ in the time-series decay factor τ(t).

[0127] Adjusting the state gain coefficient in the device state gain factor α(s) Agent's actions The calculation process that directly affects the final anomaly score A(v,t+1) at the next time step.

[0128] (III) Reward Function Design

[0129] in: Accuracy: Correctness of root cause localization (0 / 1 binary variable) False Alarm: Number of false alarms Response Time: The delay from when an exception occurs to when it is located. (iv) Policy Network Update Optimizing the strategy network using the PPO algorithm :

[0130] The dominance function Estimated using the GAE method.

[0131] 5.6 Mechanism of reinforcement learning applied to abnormal scores during reasoning During the model deployment phase, the trained policy network... The process of applying real-time parameter adjustments to the calculation of outlier scores is as follows: (I) State Awareness: The system constructs a state vector based on the current observations:

[0132] Where V represents the set of network nodes, The baseline anomaly score is calculated using the parameters from the previous time step. N is the baseline anomaly score dimension, E is the topological adjacency information dimension, and M is the historical record information dimension.

[0133] (II) Dynamic Parameter Adjustment: Policy Network according to Generate adjustment actions

[0134] in, Used to adjust the weights of multi-indicator fusion. Integer timing decay parameters, Adjust parameters related to equipment status. To ensure the effectiveness of weights and the rationality of parameters, set constraints as follows: ,and And immediately update the calculation parameters:

[0135] (III) Recalculation of outlier scores: Generating the final optimized scores using new parameters.

[0136] in, The activation function is expressed as follows: k is a user-defined coefficient used to adjust the sensitivity of the function; Piecewise exponential decay is employed:

[0137] As a key time threshold, These are topological coefficients associated with node v. This is an indicator function, indicating the device state. It takes effect when a specific condition (value 1) is met.

[0138] Please refer to Figure 4 The method in this embodiment may include: after obtaining the final anomaly score and topological influence coefficient, including: Step S401: Obtain edge weights based on the final anomaly score and topological influence coefficient; Step S402: Process the edge weights based on the preset random walk strategy to obtain the root cause probability.

[0139] Specifically, 6. Random walk algorithm and anomaly root cause path reasoning.

[0140] The root cause localization method proposed in this patent combines random walk algorithm and graph path reasoning technology. It takes the final anomaly score A(v,t) as the core input, mines potential root cause nodes through multi-level walk strategy, and realizes the interpretability localization of root causes based on anomaly propagation path analysis.

[0141] 6.1 Multi-order random walks and root cause probability modeling Based on the final anomaly score A(v,t) and its derived topological influence γ(v), the weights of edges between nodes are calculated (e.g., w(v,u) ∝ γ(u)). A(u,t)).

[0142] Please refer to Figure 5 The method in this embodiment may include: processing edge weights based on a preset random walk strategy, including: the preset random walk strategy includes at least: a first-order random walk and a second-order random walk; Step S501: First-order random walk: Generate the next hop probability based on the edge weights; Step S502: Second-order random walk: Introduce a historical node memory mechanism to balance the transition weights of the current and previous nodes through the first preset parameter.

[0143] Specifically, for the topological graph, a hybrid first-order and second-order random walk strategy is adopted: First-order random walk: Centered on the current node, the probability of the next hop is determined by the edge weight w(v_current, v_next), which is suitable for locally related scenarios.

[0144] Second-order random walk: Introducing a historical node memory mechanism, the transition weights of the current and previous nodes are balanced by the parameter β.

[0145] By generating node access trajectories through multiple rounds of walking, the access frequency is statistically analyzed and normalized into root cause probability scores, providing a candidate set for path reasoning.

[0146] Please refer to Figure 6 The method in this embodiment may include: processing the edge weights based on a preset random walk strategy, including: Step S601: Obtain a directed graph based on the root cause probability and the final anomaly score; Step S602: Process the directed graph based on Dijkstra's inverse search algorithm to obtain the localization result.

[0147] Specifically, 6.2 Weighted Reverse Path Reasoning Based on the root cause probability from the random walk, a directed graph is constructed again using A(v,t) (where the node weight S(v) is positively correlated with the root cause probability and / or A(v,t)), and an improved Dijkstra's backward search algorithm is employed: Path weight function: The reciprocal of the anomaly weight S(v_i) of node v_i in the path is used as the edge weight, ensuring that high-probability root cause nodes are more likely to be included.

[0148] S(v) is usually directly taken as A(v,t) or its combination with the root cause probability.

[0149] Termination condition: The path terminates when the cumulative weight is lower than the preset threshold or the maximum backtracking step size is reached, and the root cause node and propagation path are output.

[0150] 6.3 Core Algorithm Drivers and Results The core driving factor of the entire algorithm is A(v,t), which determines the direction of the random walk and the path cost of the reverse search. Experiments show that: The second-order walk strategy (when β=0.7) improves the root cause recall by 23.6% compared to the pure random walk.

[0151] Reverse weighted search improves the accuracy of critical path localization to over 89.2%.

[0152] Please refer to Figure 7 The method in this embodiment may include: collecting data from key devices such as gateways and set-top boxes, performing feature extraction and causal analysis, and constructing a dynamic causal network. After optimizing the causal relationships by combining domain knowledge, an anomaly scoring model is defined, and a random walk algorithm is used to track the anomaly propagation path, accurately locate the root cause node, and improve troubleshooting efficiency.

[0153] Compared with existing technologies, the advantages of this solution are: More comprehensive anomaly detection capabilities: By integrating multi-dimensional indicators, temporal changes, and network topology information, this proposal can more accurately capture anomalies in complex network environments, reduce false alarms and false negatives, and improve the accuracy and comprehensiveness of anomaly detection.

[0154] Dynamic adaptability: By employing a sliding window and reinforcement learning mechanism, the model can adapt to changes in network state, dynamically adjust parameters, maintain efficient and accurate anomaly detection performance, and eliminate the need for frequent manual intervention.

[0155] Enhanced interpretability: By using reverse weighted path reasoning, this proposal not only locates the root cause of anomalies but also provides the propagation path of anomalies, enhancing the interpretability of the results and helping operations and maintenance personnel to quickly understand and respond to network problems.

[0156] Efficiency and scalability: The algorithm design takes into account computational efficiency and resource consumption, making it suitable for large-scale network environments. At the same time, the model structure and parameters can be flexibly adjusted according to actual needs, exhibiting good scalability.

[0157] Please refer to Figure 8 The apparatus of this embodiment may include the following structure: Acquisition unit 801 is used to acquire network data sources; Analysis unit 802 is used to perform a first preprocessing on the network data source to obtain repair data; perform time granularity unification processing on the repair data to obtain unified data; perform multi-source data aggregation on the unified data to obtain associated data; process the associated data based on the time dimension algorithm to obtain a first causal graph; and process the associated data based on the static dimension algorithm to obtain a second causal graph. The processing unit 803 is used to obtain a third causal graph based on the first causal graph and the second causal graph, and to locate poor indoor network quality based on the third causal graph.

[0158] Please refer to Figure 9This embodiment of the present application also discloses an electronic device, which includes at least one processor 701, and at least one memory 702 and a bus 703 connected to the processor 701; wherein the processor 701 and the memory 702 communicate with each other through the bus 703; the processor 701 is used to call program instructions in the memory 702 to execute the above-mentioned indoor network poor quality positioning method.

[0159] The above description is merely a specific embodiment 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 technical scope disclosed in 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 method for locating locations with poor indoor network quality, characterized in that, The method includes: Obtain network data source; The network data source is preprocessed to obtain repaired data; The repaired data is processed to unify the time granularity to obtain unified data; The unified data is aggregated from multiple sources to obtain related data; The associated data is processed using a time-dimensional algorithm to obtain a first causal graph; The associated data is processed using a static dimensionality algorithm to obtain a second causal graph; Based on the first causal graph and the second causal graph, a third causal graph is obtained, and the poor indoor network quality is located based on the third causal graph.

2. The method according to claim 1, characterized in that, The unified data is aggregated from multiple sources to obtain related data, including: The unified data includes at least: device ID and timestamp; The associated data is obtained based on the device ID and the timestamp.

3. The method according to claim 2, characterized in that, The process of processing the associated data using a time-dimensional algorithm to obtain a first causal graph includes: The associated data is processed using the No Tears algorithm to obtain a first causal graph.

4. The method according to claim 3, characterized in that, The step of processing the associated data based on the static dimensionality algorithm to obtain the second causal graph includes: The associated data is processed using the TTPM algorithm to obtain a second causal graph.

5. The method according to claim 4, characterized in that, Locating poor indoor network quality based on the third cause-effect graph includes: The third causal graph is processed based on the first preset strategy to obtain the final anomaly score and topological influence coefficient.

6. The method according to claim 5, characterized in that, After obtaining the final anomaly score and topological influence coefficient, the process includes: The edge weights are obtained based on the final anomaly score and the topological influence coefficient. The edge weights are processed based on a preset random walk strategy to obtain the root cause probability.

7. The method according to claim 6, characterized in that, The edge weights are processed based on a preset random walk strategy, including: The preset random walk strategy includes at least: first-order random walk and second-order random walk; The first-order random walk generates the next hop probability based on the edge weights; The second-order random walk introduces a historical node memory mechanism, which balances the transfer weights of the current and previous nodes through a first preset parameter.

8. The method according to claim 7, characterized in that, After processing the edge weights based on a preset random walk strategy, the process includes: Based on the root cause probability and the final anomaly score, a directed graph is obtained; The directed graph is processed using Dijkstra's inverse search algorithm to obtain the localization result.

9. A positioning device, characterized in that, The device includes: an acquisition unit, an analysis unit, and a processing unit. The acquisition unit is used to acquire network data sources; The analysis unit is configured to perform a first preprocessing on the network data source to obtain repair data; perform time granularity unification processing on the repair data to obtain unified data; perform multi-source data aggregation on the unified data to obtain associated data; process the associated data based on a time-dimensional algorithm to obtain a first causal graph; and process the associated data based on a static-dimensional algorithm to obtain a second causal graph. The processing unit is configured to obtain a third causal graph based on the first causal graph and the second causal graph, and locate poor indoor network quality based on the third causal graph.

10. An electronic device, characterized in that, include: Memory, used to store at least one set of instructions; The processor is used to acquire network data sources; The network data source is preprocessed to obtain repaired data; The repaired data is processed to unify the time granularity to obtain unified data; The unified data is aggregated from multiple sources to obtain related data; The associated data is processed using a time-dimensional algorithm to obtain a first causal graph; The associated data is processed using a static dimensionality algorithm to obtain a second causal graph; Based on the first causal graph and the second causal graph, a third causal graph is obtained, and the poor indoor network quality is located based on the third causal graph.

Images