Wireless network coverage prediction method

Abstract

The invention provides a wireless network coverage prediction method, which belongs to the technical field of wireless communication, and comprises the following steps: constructing a wireless signal prediction model, setting correction factors and loss parameters according to different geographical environments, and carrying out preliminary prediction on the network coverage condition; taking the test base station as a center, actually measuring base station wireless signal indexes of different distances and ranges according to a geographical environment, and fitting actual measurement data to obtain a network coverage condition; and performing iterative correction on the correction factor and the loss parameter of the prediction model for different geographical environments by combining the prediction data and the actual measurement data of the prediction model to obtain network coverage prediction data in different geographical environments. The method has the advantages that a prediction framework based on multi-model fusion effectively deals with the influence of a complex geographical environment, signal attenuation change caused by topographic relief and building shielding can be accurately quantified, and the pain point of prediction misalignment of a traditional method in variable terrains and dense urban areas is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communications, and in particular relates to a wireless network coverage prediction method. Background Art

[0002] In modern power wireless private networks, network coverage prediction is a core requirement for ensuring service reliability and efficiency. The power industry relies on stable wireless communications to support critical applications such as smart grid monitoring and remote control. However, the complex geographical environment (including variable terrain, densely built-up areas, and diverse usage scenarios) leads to uneven signal coverage, frequent blind spots, and interference. This uncertainty directly threatens the safety and efficiency of the power system. A prediction method that provides continuous and accurate coverage assessment and dynamically adapts to environmental changes is urgently needed to meet real-time service needs while reducing manual intervention and optimizing network resources. Summary of the Invention

[0003] In view of this, the present invention addresses the deficiencies in the prior art and provides a wireless network coverage prediction method. To solve the above technical problems, the technical solution adopted by the present invention is: including: constructing a wireless signal prediction model, setting correction factors and loss parameters according to different geographical environments, and making a preliminary prediction of the network coverage; taking the test base station as the center, actually measuring the wireless signal indicators of base stations at different distances and ranges according to the geographical environment, and fitting the actual measurement data to obtain the network coverage; combining the predicted data and actual measurement data of the prediction model, iteratively correcting the correction factors and loss parameters of the prediction model for different geographical environments, and obtaining network coverage prediction data under different geographical environments; using terminal equipment to collect and report wireless signal field strength data to the existing network operation and maintenance platform, and the existing network operation and maintenance platform continuously correcting the correction factors and loss parameters of the prediction model according to the wireless signal field strength data.

[0004] Furthermore, a wireless signal prediction model is constructed, and correction factors and loss parameters are set according to different geographical environments. Methods for making preliminary predictions of network coverage include:

[0005] Based on the Okumura-Hata model, the basic free space path loss is obtained:

[0006] L b =69.55+26.16logf-13.82logh b -a(h m )+(44.9-6.55logh b )logd;

[0007] Where: f is the signal frequency; h b is the base station height; h m is the height of the terminal; d is the distance between the base station and the terminal; a(hm ) is the terminal altitude correction factor;

[0008] In an urban propagation environment, a CCIR model is constructed based on the free space basic path loss and urban environment loss. The model-corrected path loss is: L 总 =L b +ΔL 地形 +ΔL 建筑 ; Generate a terrain correction factor by measuring the altitude difference and slope change terrain characteristics of the area, expressed as ΔL 地形 =k1·log(1+σ alt )+k2·θ slope ; Based on artificial building coverage and average height H avg And the building material type penetration coefficient β generates the building density correction factor, expressed as: ΔL 建筑 =K 建筑物 ·log(1+β·H avg );

[0009] Where: L b : Basic path loss, ΔL 地形 : Independently calculated compensation for signal attenuation caused by terrain fluctuation, σ alt is the degree of three-dimensional terrain relief; ΔL 建筑 : independently calculated compensation for signal penetration loss due to building density and structure; k1, k2: power scenario calibration coefficients determined by actual measurement; θ slope : Maximum slope angle; K 建筑物 is the building density weight;

[0010] Based on the line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios, the COST231-WI model is constructed:

[0011] LOS path loss: L LOS =42.6+26logd+20logf;

[0012] Added diffraction compensation item for NLOS scenarios:

[0013]

[0014] Where: ΔL 绕射 and L 多径 is the correction for diffraction loss and multipath fading; φ is the diffraction angle of the obstacle; τrms is the delay spread, which is extracted from the channel sounding signal.

[0015] Furthermore, with the test base station as the center, the wireless signal indicators of base stations at different distances and ranges are actually measured according to the geographical environment, and the method of fitting the actual measurement data to obtain the network coverage includes:

[0016] Select multiple test sites in the deployment area, set up test base stations at the test sites, and measure and collect the latitude and longitude coordinates, RSSI values, and SNR values ​​of the test sites;

[0017] Use the Kriging interpolation algorithm to fit discrete data and generate signals that continuously cover the thermal range; verify data stability and eliminate outliers;

[0018] Evaluate the wireless signal coverage of the test base station and use it to estimate the coverage of the same or similar base stations;

[0019] Use the ray method or loop method for dynamic drive testing to improve the accuracy and coverage of data collection.

[0020] Furthermore, the methods for generating continuous signals covering the thermal range by fitting discrete data using the Kriging interpolation algorithm include:

[0021] Calculate the semivariogram function value γ(h) between all measurement points and establish a function model by distance grouping

[0022] The target area is divided into 10m×10m grid cells, and the weight coefficient λ is solved based on the semivariogram function for each grid center point s0. i , through the weighted average formula Predicted signal strength value;

[0023] Mapping the predicted signal strength value of the grid unit to a continuous color scale;

[0024] Where: h represents the distance between measurement points; N(h) is the number of point pairs with a specific distance h; Z(s i ) and Z(s i +h) are respectively the position s i and s i The measured signal strength at point +h is the RSSI value collected by the ray method / loop method; Z(s0) is the predicted signal strength at the grid center point s0.

[0025] Further, methods for verifying data stability and removing outliers include:

[0026] Calculate the interpolation residual δ of each measurement point i =|Z pred (s i )-Z meas (s i )∣, when δ i When >3σ, the outlier discrimination mechanism is activated:

[0027] If the point's environment tag contains "building occlusion" or "terrain abrupt change," retain the data and add a compensation tag;

[0028] Otherwise, mark it as an outlier and use interpolation Replacement;

[0029] Where: i is the interpolation residual; σ is the standard deviation of the regional residual, σ≤5dB; Z pred (s i ) is the signal strength predicted by Kriging interpolation; Z meas (s i ) is at position s i Actual measured signal strength; Z rep (s i ) is the position s i The replacement value at s i is the measuring point; s j is the neighborhood measurement point; φ(·) is the Gaussian kernel function, which controls the spatial attenuation weight, φ(r)=e -σr2 ;||s i -s j || is s i With s j The distance between them.

[0030] Furthermore, combining the prediction data of the prediction model with the actual measurement data, iteratively correcting the correction factor and loss parameter of the prediction model for different geographical environments, and obtaining network coverage prediction data in different geographical environments include:

[0031] Assume that the model-predicted field strength is E-mode and the measured field strength is E-real, and calculate the deviation ΔE=|E-mode-E-real|;

[0032] If ΔE>6dB, adjust the correction factor according to the gradient descent method. Correction ends when max(ΔE)≤3dB in the whole area;

[0033] Where: Δβ is the change in the correction factor β in a single iteration, which controls the parameter optimization step size; η is the learning rate, a hyperparameter that controls the convergence speed. A large value will cause oscillation, while a small value will slow convergence. It is the partial derivative of the deviation with respect to the correction factor, reflecting the sensitivity of the field intensity deviation ΔE to the change of β, and is used to determine the optimization direction;

[0034] Establish correction factor parameter libraries for three scenarios: mountainous or urban terrain, building height ≤15m or >30m, and high business density or low business density, and obtain network coverage prediction data based on the correction factor parameter libraries.

[0035] Furthermore, the method of using terminal equipment to collect and report wireless signal field strength data to the existing network operation and maintenance platform includes:

[0036] The terminal device collects the downlink signal RSSI value, SNR value, GPS positioning coordinates and timestamp at a period T, where the collection period T satisfies: When RSSI≤-95dBm or SNR≤10dB is detected, real-time reporting is triggered immediately;

[0037] Abnormal data screening criteria: Eliminate data points with GPS positioning deviation greater than 50 meters; add an "environmental interference" mark to data with RSSI mutation greater than 20dB between adjacent reporting points;

[0038] Aggregate data by geographic grid, and perform weighted average of terminal data within the same grid:

[0039]

[0040] Furthermore, the method for the existing network operation and maintenance platform to continuously correct the correction factor and loss parameter of the prediction model based on the wireless signal field strength data includes:

[0041] Deviation driven correction: Where: N ij is the number of terminals in the grid (i, j); η is the learning rate, which controls the parameter update step size, η = 0.1; Δβ ij is the adjustment amount of the correction factor, which means the correction factor β required for the grid (i, j) ij the amount of change applied; is the partial derivative operator, which quantifies the prediction deviation to β ij Sensitivity; N ij is the number of terminal measurement points that meet the quality requirements within the grid (i, j); is the predicted signal strength at position k based on the propagation mode; is the average RSSI actually measured by the terminal at position k;

[0042] When the cumulative data volume of a single grid is greater than 500 groups, a grid-specific correction factor β is generated. ij ; When Δβ>0.05 for 10 consecutive times, a new scene parameter group is created;

[0043] Add compensation item for rain and fog weather data: α new =α old +0.03×rainfall intensity; where the unit of rainfall intensity is mm / h.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. This application's prediction framework, based on multi-model fusion, effectively addresses the impact of complex geographical environments. It can accurately quantify changes in signal attenuation caused by terrain undulations and building obstructions, resolving the pain points of traditional methods inaccurate predictions in variable terrain and densely populated urban areas. This predictive capability enables more scientific network planning to avoid signal blind spots, providing reliable support for critical services such as power wireless private networks.

[0046] 2. In terms of data processing, innovative anomaly identification and compensation strategies significantly improve data reliability. Through a dual verification mechanism of residual analysis and environmental labeling, the system intelligently distinguishes between true anomalies and legitimate deviations caused by environmental factors. Data reconstruction technology based on spatial correlation effectively maintains geographic continuity while eliminating invalid data, providing a cleaner input data source for predictive models.

[0047] 3. The dynamic correction mechanism of the present invention gives the prediction model the ability to continuously evolve. Driven by the measured data reported by the terminal, the system can automatically adjust the correction factor and loss parameters to adapt to environmental changes in different scenarios in real time; the specially established scenario parameter library can intelligently match diverse landform features such as mountainous areas and urban areas. Combined with the weather compensation mechanism, it greatly improves the environmental adaptability and timeliness of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be described in further detail below with reference to the accompanying drawings.

[0049] Figure 1 : Schematic diagram of the process of the present invention; DETAILED DESCRIPTION

[0050] In order to better understand the present invention, the content of the present invention is further clearly described below in conjunction with the examples and drawings, but the protection content of the present invention is not limited to the following examples. In the following description, a large number of specific details are given in order to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details.

[0051] Example 1: See Figure 1 , a wireless network coverage prediction method of this embodiment includes:

[0052] S1. Build a wireless signal prediction model, set correction factors and loss parameters based on different geographical environments, and make preliminary predictions of network coverage. Implement environmentally adaptive predictions through multi-model fusion based on the Okumura-Hata model, the physical basis of electromagnetic wave propagation. A gradient prediction framework is formed through a three-pronged collaborative mechanism—a basic propagation model, an environmental correction model, and a scenario enhancement model. This framework maintains the universality of the physical model while ensuring physical accuracy, improving the reliability of coverage predictions in complex power scenarios.

[0053] Wireless signal prediction models include:

[0054] The free space basic path loss L is calculated using the Okumura-Hata model. b =69.55+26.16logf-13.82logh b -a(h m )+(44.9-6.55logh b )logd;

[0055] The terminal height correction factor a(h m ) is used to compensate for differences in signal reception caused by differences in the deployment heights of different terminals;

[0056] Where: f is the signal frequency; h b is the base station height; h m is the height of the terminal; d is the distance between the base station and the terminal; a(h m ) is the terminal altitude correction factor.

[0057] In an urban propagation environment, a CCIR model is constructed based on the free space basic path loss and urban environment loss. The model-corrected path loss is: L 总 =L b +ΔL 地形 +ΔL 建筑 The dual-factor loss compensation mechanism avoids interference from terrain and buildings, maintaining correction independence in complex scenarios such as hilly urban areas.

[0058] By measuring the altitude difference and slope change of the measured area, a terrain correction factor is generated, which can accurately predict the signal attenuation of typical power corridors such as valleys and ridges, expressed as ΔL 地形 =k1·log(1+σ alt )+k2·θ slope ; Based on artificial building coverage and average height H avg And the building material type penetration coefficient β generates the building density correction factor, expressed as: ΔL 建筑 =K 建筑物 ·log(1+β·H avg ); β parameter differentiates the penetration characteristics of concrete / glass building materials; effectively distinguishes the coverage differences between commercial areas and industrial areas in dense urban areas;

[0059] Where: L b : Basic path loss, ΔL 地形 : independently calculated compensation for signal attenuation caused by terrain undulation; σ alt Quantify the three-dimensional terrain relief; ΔL 建筑: independently calculated compensation for signal penetration loss due to building density and structure; k1, k2: power scenario calibration coefficients determined by actual measurement; θ slope : Maximum slope angle; K 建筑物 is the building density weight;

[0060] Based on the line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios, the COST231-WI model is constructed. The scenario-based modeling mechanism dynamically adapts to changes in the propagation path and solves the problem of sudden line-of-sight changes when the transmission line crosses a mountain. The COST231-WI model is expressed as:

[0061] LOS path loss: L LOS =42.6+26logd+20logf;

[0062] Added diffraction compensation item for NLOS scenario:

[0063] The multipath term quantifies the reflection effect of the metal structure, and the diffraction term compensates for the shielding problem of the transformer metal building group, improving the prediction stability in the complex electromagnetic environment of the substation;

[0064] Where: ΔL 绕射 and L 多径 is the correction for diffraction loss and multipath fading; φ is the diffraction angle of the obstacle; τrms is the delay spread, which is extracted from the channel sounding signal.

[0065] S2. Focusing on the test base station, we measure base station wireless signal indicators at different distances and ranges based on the actual geographical environment, and fit the actual measured discrete data to obtain network coverage. We also construct a real-world propagation model through spatial discrete sampling to obtain first-hand data on the actual electromagnetic environment, including:

[0066] Multiple test sites were selected within the deployment area, and a multi-base station collaborative sampling strategy was used to increase the test quantity benchmark and address measurement challenges in hilly areas with large signal fluctuations. Test base stations were set up at the test sites to measure and collect the latitude and longitude coordinates, RSSI values, and SNR values.

[0067] By fitting discrete data using the Kriging interpolation algorithm, a signal continuously covers the thermal range. Based on the interpolation principle of spatial autocorrelation, the signal attenuation gradient is quantified, and a continuous coverage map that conforms to the laws of electromagnetic wave propagation is generated, making the data more intuitive. Data stability is verified, outliers are eliminated, and environmental interference is analyzed and separated from true anomalies to ensure data reliability in strong electromagnetic environments such as substations.

[0068] Evaluate the wireless signal coverage of the test base station and use this to estimate the coverage of the same or similar base stations. By clustering base station features, knowledge transfer can be achieved, reducing the workload of repeated drive tests in areas with dense power base stations.

[0069] Use the ray method or loop method for dynamic road testing to improve the accuracy and coverage of data collection. The ray method is suitable for complex terrain, and the loop method optimizes the sampling density in urban areas.

[0070] By fitting discrete data using the Kriging interpolation algorithm, a continuous signal is generated covering the thermal range, including:

[0071] The semivariance function γ(h) quantifies the spatial autocorrelation of signal strength, which can objectively reflect the nonlinear characteristics of electromagnetic wave attenuation with distance, overcome the distortion problem of simple linear interpolation, calculate the semivariance function value γ(h) between all measurement points, and establish a function model by distance grouping. The distance-based grouping strategy adapts to changing terrain environments and maintains computational stability in complex scenarios such as power transmission lines in mountainous areas.

[0072] The target area is divided into 10m×10m grid cells. For each grid center point s0, the weight coefficient λ is solved based on the semivariogram function. i , through the weighted average formula Predict signal strength values; the spatial weighting mechanism strengthens the influence weight of neighboring points and accurately restores the signal strength mutation characteristics in the base station near-field area;

[0073] The predicted signal strength values ​​of grid cells are mapped into continuous color scales. This color scale mapping uses a color spectrum scheme that is easily distinguished by the human eye, visually presenting coverage blind spots and interference areas, helping to improve operation and maintenance decision-making efficiency.

[0074] Where: h represents the distance between measurement points; N(h) is the number of point pairs with a specific distance h; Z(s i ) and Z(s i +h) are respectively the position s i and s i The measured signal strength at point +h, the RSSI value collected by the ray method / loop method; Z(s0) is the predicted signal strength at the grid center point s0

[0075] This step uses the Kriging algorithm to model spatial correlation, restore the real electromagnetic environment characteristics with a 10-meter grid accuracy, and provide a coverage heat map with millimeter-level accuracy for the power wireless private network.

[0076] Verify data stability, eliminate outliers, and use intelligent discrimination mechanisms to maximize environmental feature information while ensuring data quality, providing highly reliable data for model correction, including:

[0077] The interpolation residual calculation mechanism establishes a basis for data reliability assessment by quantifying the deviation between the predicted value and the measured value. It can effectively identify abnormal data caused by non-environmental factors, improve the accuracy of coverage prediction, and calculate the interpolation residual δ of each measurement point.i =|Z pred (s i )-Z meas (s i )∣, when δ i When the value is greater than 3σ, the outlier discrimination mechanism is activated. The 3σ threshold covers the normal data fluctuation range based on statistical principles. While ensuring data integrity, it significantly reduces the misjudgment rate. The outlier discrimination mechanism is as follows:

[0078] If the point's environmental tag contains "building occlusion" or "terrain mutation," the data is retained and a compensation tag is added. The environmental tag mechanism retains reasonable attenuation data for model compensation to avoid loss of valid information.

[0079] Otherwise, mark it as an outlier and use interpolation Substitution; Gaussian kernel interpolation uses spatial correlation to reconstruct data, maintaining geographic continuity while removing outliers;

[0080] Where: i is the interpolation residual; σ is the standard deviation of the regional residual. In this embodiment, σ≤5dB. The standard deviation threshold is set with reference to the wireless signal measurement error range to adapt to the signal fluctuation characteristics in different environments. pred (s i ) is the signal strength predicted by Kriging interpolation; Z meas (s i ) is at position s i Actual measured signal strength; Z rep (s i ) is the position s i The replacement value at s i is the measuring point; s j is the neighborhood measurement point; φ(·) is the Gaussian kernel function, which controls the spatial attenuation weight. The exponential attenuation form conforms to the propagation characteristics of electromagnetic waves, ensuring that the closer the distance, the higher the weight of the measurement point; φ(r)=e -σr2 ;||s i -s j || is s i With s j The distance between them.

[0081] S3. Combining the prediction model's predicted data with actual measurement data, iteratively adjust the prediction model's correction factors and link loss parameters for different geographical environments to obtain network coverage prediction data for different geographical environments. A closed-loop feedback mechanism is used to achieve self-optimization of the prediction model, dynamically adapting to environmental changes and addressing the accuracy degradation problem of traditional models after long-term use. Specific steps include:

[0082] Assume that the model-predicted field strength is E-mode and the measured field strength is E-real, and calculate the deviation ΔE=|E-mode-E-real|;

[0083] If ΔE>6dB, the correction factor is adjusted according to the gradient descent method. The 6dB threshold is set to balance the response speed and system stability, avoid small fluctuations triggering invalid corrections, and reduce computing resource consumption. Correction is terminated until max(ΔE)≤3dB in the entire area. Gradient descent method is used to automatically optimize parameters, replacing manual trial and error to improve parameter optimization efficiency.

[0084] Where: Δβ is the change in the correction factor β in a single iteration, which controls the parameter optimization step size; η is the learning rate, a hyperparameter that controls the convergence speed. If the value is too large, it will cause oscillation, and if it is too small, it will converge slowly. The learning rate η establishes a balance between convergence speed and stability, preventing oscillation and divergence in the optimization process and ensuring system robustness. It is the partial derivative of the deviation with respect to the correction factor, reflecting the sensitivity of the field intensity deviation ΔE to the change of β, and is used to determine the optimization direction;

[0085] Establish a correction factor parameter library for three scenarios: mountainous or urban terrain, building height ≤15m or >30m, and high business density or low business density. The parameter library enables knowledge accumulation and rapid matching of scenarios, and obtains network coverage prediction data based on the correction factor parameter library.

[0086] Technical effects of this embodiment:

[0087] 1. This application's prediction framework, based on multi-model fusion, effectively addresses the impact of complex geographical environments. It can accurately quantify changes in signal attenuation caused by terrain undulations and building obstructions, resolving the pain points of traditional methods inaccurate predictions in variable terrain and densely populated urban areas. This predictive capability enables more scientific network planning to avoid signal blind spots, providing reliable support for critical services such as power wireless private networks.

[0088] 2. In terms of data processing, innovative anomaly identification and compensation strategies significantly improve data reliability. Through a dual verification mechanism of residual analysis and environmental labeling, the system intelligently distinguishes between true anomalies and legitimate deviations caused by environmental factors. Data reconstruction technology based on spatial correlation effectively maintains geographic continuity while eliminating invalid data, providing a cleaner input data source for predictive models.

[0089] Example 2: Based on a wireless network coverage prediction method implemented as in Example 1, refer to Figure 1 , also includes:

[0090] S4. Use terminal devices to collect and report wireless signal strength data to the existing network operation and maintenance platform, establish a dynamic strategy for terminal-side data collection, balance data timeliness and system energy consumption, and achieve lightweight operation of smart grid terminals. Specific steps include:

[0091] The terminal device collects the downlink signal RSSI value, SNR value, GPS positioning coordinates and timestamp at a period T, where the collection period T satisfies: The cycle dynamic adjustment mechanism responds to changes in business load, with intensive data collection during peak periods to ensure key business monitoring, and extended cycles during off-peak periods to reduce terminal energy consumption;

[0092] When RSSI ≤ -95dBm or SNR ≤ 10dB is detected, real-time reporting is triggered immediately. The weak coverage threshold trigger mechanism captures network anomalies and provides real-time warning of communication interruption risks at key nodes such as substations.

[0093] Abnormal data screening criteria: Data points with GPS positioning deviations greater than 50 meters are eliminated. Positioning accuracy filtering ensures geographic reliability and avoids data drift in linear deployment scenarios such as transmission lines. "Environmental interference" flags are added to data where RSSI changes greater than 20dB between adjacent reporting points. The mutation flagging mechanism distinguishes between true coverage changes and transient interference, and identifies strong electromagnetic environments such as high-voltage transmission lines, ensuring data validity.

[0094] Aggregate data by geographic grid, and perform weighted average of terminal data within the same grid: The weight grading mechanism strengthens the voice of key terminals and prioritizes the accuracy of data for key businesses such as smart meters and relay protection.

[0095] S5. The existing network operation and maintenance platform continuously adjusts the correction factors and loss parameters of the prediction model based on wireless signal strength data, establishing a closed-loop "prediction-measurement-correction" system. This enables the model to continuously evolve and prevents the prediction model from becoming rigid. Specific steps include:

[0096] Deviation driven correction: Where: N ij is the number of terminals in the grid (i, j). Grid processing enables refined spatial management and precise coverage optimization in key areas such as substations and transmission lines. η is the learning rate, which controls the parameter update step size. η = 0.1. By balancing the convergence speed and stability with a fixed learning rate, we can avoid optimization oscillations caused by signal fluctuations in hilly areas. Δβ ij is the adjustment amount of the correction factor, which means the correction factor β required for the grid (i, j) ij The applied variation and adjustment amount directly act on the correction factor to achieve gradient optimization, significantly improving the calibration efficiency of the penetration coefficient of buildings in dense urban areas; is the partial derivative operator, which quantifies the prediction deviation to β ij Sensitivity; N ij The number of terminal measurement points that meet the quality requirements within the grid (i, j) is used. The sensitivity analysis identifies the key influencing parameters and prioritizes the optimization of the correction factors that have the greatest impact on coverage quality. is the predicted signal strength at position k based on the propagation mode; is the average RSSI actually measured by the terminal at position k; |·| represents the absolute value of the predicted value and the measured value;

[0097] When the cumulative data volume of a single grid is greater than 500 groups, a grid-specific correction factor β is generated. ij , the data volume threshold ensures statistical significance and provides customized parameters for different scenarios; when Δβ>0.05 for 10 consecutive times, a new scenario parameter group is created, and the parameter mutation detection mechanism responds to sudden environmental changes, promptly adapting to sudden events such as landslides in mountainous areas and new high-rise buildings in urban areas;

[0098] Add compensation item for rain and fog weather data: α new =α old +0.03× rainfall intensity; where: rainfall intensity is in mm / h; dynamically compensates for meteorological attenuation effects to improve forecast reliability in severe weather such as heavy rain.

[0099] Technical effects of this embodiment

[0100] 1. The dynamic correction mechanism of the present invention gives the prediction model the ability to continuously evolve. Driven by the measured data reported by the terminal, the system can automatically adjust the correction factor and loss parameters to adapt to environmental changes in different scenarios in real time. The specially established scenario parameter library can intelligently match the diverse landform features such as mountainous areas and urban areas. Combined with the weather compensation mechanism, it greatly improves the environmental adaptability and timeliness of the prediction model.

[0101] 2. This invention significantly optimizes resource efficiency while ensuring prediction accuracy. The intelligently triggered data reporting mechanism significantly reduces network signaling overhead, and geographic grid aggregation technology effectively compresses data processing. This overall solution reduces reliance on manual drive testing, making network coverage optimization more efficient and economical, and providing strong communication support for critical infrastructure such as smart grids.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limiting. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A wireless network coverage prediction method, characterized in that: include: Build a wireless signal prediction model, set correction factors and loss parameters based on different geographical environments, and make preliminary predictions on network coverage; Centered on the test base station, the wireless signal indicators of base stations at different distances and ranges are measured according to the geographical environment, and the network coverage is obtained by fitting the actual measurement data. Combining the prediction model's prediction data with actual measurement data, the correction factors and loss parameters of the prediction model are iteratively modified for different geographical environments to obtain network coverage prediction data in different geographical environments. The terminal equipment is used to collect and report wireless signal field strength data to the existing network operation and maintenance platform, and the existing network operation and maintenance platform continuously adjusts the correction factor and loss parameter of the prediction model according to the wireless signal field strength data.

2. The wireless network coverage prediction method according to claim 1, wherein: Methods for constructing a wireless signal prediction model, setting correction factors and loss parameters based on different geographical environments, and making preliminary predictions about network coverage include: Based on the Okumura-Hata model, the basic free space path loss is obtained: L b =69.55+26.16logf-13.82logh b -a(h m )+(44.9-6.55logh b )logd; Where: f is the signal frequency; h b is the base station height; h m is the height of the terminal; d is the distance between the base station and the terminal; a(h m ) is the terminal altitude correction factor; In an urban propagation environment, a CCIR model is constructed based on the free space basic path loss and urban environment loss. The model-corrected path loss is: L 总 =L b +ΔL 地形 +ΔL 建筑 ; Generate a terrain correction factor by measuring the altitude difference and slope change terrain characteristics of the area, expressed as ΔL 地形 =k1·log(1+σ alt )+k2·θ slope ; Based on artificial building coverage and average height H avg And the building material type penetration coefficient β generates the building density correction factor, expressed as: ΔL 建筑 =K 建筑物 ·log(1+β·H avg ); Where: L b : Basic path loss, ΔL 地形 : Independently calculated compensation for signal attenuation caused by terrain fluctuation, σ alt is the degree of three-dimensional terrain relief; ΔL 建筑 : independently calculated compensation for signal penetration loss due to building density and structure; k1, k2: power scenario calibration coefficients determined by actual measurement; θ slope : Maximum slope angle; K 建筑物 is the building density weight; Based on the line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios, the COST231-WI model is constructed: LOS path loss: L LOS =42.6+26logd+20logf; Added diffraction compensation item for NLOS scenarios: Where: ΔL 绕射 and L 多径 is the correction for diffraction loss and multipath fading; φ is the diffraction angle of the obstacle; τrms is the delay spread, which is extracted from the channel sounding signal.

3. The wireless network coverage prediction method according to claim 1, wherein: Centered on the test base station, the wireless signal indicators of base stations at different distances and ranges are measured according to the geographical environment. The methods for fitting the actual measurement data to obtain network coverage include: Select multiple test sites in the deployment area, set up test base stations at the test sites, and measure and collect the latitude and longitude coordinates, RSSI values, and SNR values ​​of the test sites; Use the Kriging interpolation algorithm to fit discrete data and generate signals that continuously cover the thermal range; verify data stability and eliminate outliers; Evaluate the wireless signal coverage of the test base station and use it to estimate the coverage of the same or similar base stations; Use the ray method or loop method for dynamic drive testing to improve the accuracy and coverage of data collection.

4. The wireless network coverage prediction method according to claim 3, wherein: Methods for generating continuous signals covering the thermal range by fitting discrete data using the Kriging interpolation algorithm include: Calculate the semivariogram function value γ(h) between all measurement points and establish a function model by distance grouping The target area is divided into 10m×10m grid cells, and the weight coefficient λ is solved based on the semivariogram function for each grid center point s0. i , through the weighted average formula Predicted signal strength value; Mapping the predicted signal strength value of the grid unit to a continuous color scale; Where: h represents the distance between measurement points; N(h) is the number of point pairs with a specific distance h; Z(s i ) and Z(s i +h) are respectively the position s i and s i The measured signal strength at point +h is the RSSI value collected by the ray method / loop method; Z(s0) is the predicted signal strength at the grid center point s0.

5. The wireless network coverage prediction method according to claim 4, wherein: Methods for verifying data stability and removing outliers include: Calculate the interpolation residual δ of each measurement point i =|Z pred (s i )-Z meas (s i )∣, when δ i When >3σ, the outlier discrimination mechanism is activated: If the point's environment tag contains "building occlusion" or "terrain abruptness," retain the data and add a compensation mark; Otherwise, mark it as an outlier and use interpolation Replacement; Where: i is the interpolation residual; σ is the standard deviation of the regional residual, σ≤5dB; Z pred (s i ) is the signal strength predicted by Kriging interpolation; Z meas (s i ) is at position s i Actual measured signal strength; Z rep (s i ) is the position s i The replacement value at s i is the measuring point; s j is the neighborhood measurement point; φ(·) is the Gaussian kernel function, which controls the spatial attenuation weight, φ(r)=e -σr2 ;||s i -s j || is s i With s j The distance between them.

6. The wireless network coverage prediction method according to claim 1, wherein: Combining the prediction data of the prediction model with actual measurement data, iteratively correcting the correction factors and loss parameters of the prediction model for different geographical environments. Methods for obtaining network coverage prediction data in different geographical environments include: Assume that the model-predicted field strength is E-mode and the measured field strength is E-real, and calculate the deviation ΔE=|E-mode-E-real|; If ΔE>6dB, adjust the correction factor according to the gradient descent method. Correction ends when max(ΔE)≤3dB in the whole area; Where: Δβ is the change in the correction factor β in a single iteration, which controls the parameter optimization step size; η is the learning rate, a hyperparameter that controls the convergence speed. A large value will cause oscillation, while a small value will slow convergence. It is the partial derivative of the deviation with respect to the correction factor, reflecting the sensitivity of the field intensity deviation ΔE to the change of β, and is used to determine the optimization direction; Establish correction factor parameter libraries for three scenarios: mountainous or urban terrain, building height ≤15m or >30m, and high business density or low business density, and obtain network coverage prediction data based on the correction factor parameter libraries.

7. The wireless network coverage prediction method according to claim 1, wherein: Methods for using terminal devices to collect and report wireless signal strength data to the existing network operation and maintenance platform include: The terminal device collects the downlink signal RSSI value, SNR value, GPS positioning coordinates and timestamp at a period T, where the collection period T satisfies: When RSSI≤-95dBm or SNR≤10dB is detected, real-time reporting is triggered immediately; Abnormal data screening criteria: Eliminate data points with GPS positioning deviation greater than 50 meters; add an "environmental interference" mark to data with RSSI mutation greater than 20dB between adjacent reporting points; Aggregate data by geographic grid, and perform weighted average of terminal data within the same grid:

8. The wireless network coverage prediction method according to claim 7, wherein: The method for the existing network operation and maintenance platform to continuously correct the correction factor and loss parameter of the prediction model based on the wireless signal field strength data includes: Deviation driven correction: Where: N ij is the number of terminals in the grid (i, j); η is the learning rate, which controls the parameter update step size, η = 0.1; Δβ ij is the adjustment amount of the correction factor, which means the correction factor β required for the grid (i, j) ij the amount of change applied; is the partial derivative operator, which quantifies the prediction deviation to β ij Sensitivity; N ij is the number of terminal measurement points that meet the quality requirements within the grid (i, j); is the predicted signal strength at position k based on the propagation mode; is the average RSSI actually measured by the terminal at position k; When the cumulative data volume of a single grid is greater than 500 groups, a grid-specific correction factor β is generated. ij ; When Δβ>0.05 for 10 consecutive times, a new scene parameter group is created; Add compensation item for rain and fog weather data: α new =α old +0.03×rainfall intensity; where the unit of rainfall intensity is mm / h.

Images