4G DTU Signal Strength Identification Method Based on Edge Data Transmission

Abstract

This invention discloses a 4G DTU signal strength identification method based on edge data transmission, relating to the field of 4G DTU signal strength identification. The method includes: collecting and preprocessing edge operation data of the 4G DTU device and reading a location fingerprint database; generating a service-RF coupling feature vector after performing cross-layer coupling coding; inputting the data into an edge construction layer to generate an edge signal state vector; inputting the data into a dual-constant layer to generate a modulated signal fluctuation time constant and a modulated transmission anomaly time constant; inputting the data into a gated recursive layer to generate a signal strength state representation; performing interlock protection determination to generate an RF protective derating instruction; updating the location fingerprint database, and generating a 4G DTU signal strength identification result. This invention uses an edge LTC model to identify 4G DTU signals, possessing the advantages of accuracy, stability, and location adaptability.

Description

Technical Field

[0001] This invention relates to the field of 4G DTU signal strength identification, and more particularly to a 4G DTU signal strength identification method based on edge data transmission. Background Technology

[0002] Currently, in the field of 4G DTU edge data transmission, existing technologies typically determine communication quality by reading CSQ signal strength values ​​or network registration status, and provide indications using fixed thresholds or simple indicator light colors. This method relies heavily on a single signal indicator, making it difficult to reflect actual transmission conditions such as upload latency, retransmissions, and increased power consumption. This can easily lead to issues where the signal appears normal, but edge data uploads are unstable.

[0003] Meanwhile, existing technologies often employ fixed signal level boundaries and static display strategies during on-site installation and commissioning, lacking continuous calibration based on differences in installation locations, variations in RF power consumption, and historical transmission performance. When the 4G DTU is located in a metal cabinet, a remote site, or a weak coverage area, signal fluctuations and link degradation interact, making it difficult for traditional methods to trigger protection controls in a timely manner, and also unable to update the identification benchmark based on site operation records. Summary of the Invention

[0004] One objective of this invention is to propose a 4G DTU signal strength identification method based on edge data transmission. This invention uses an edge LTC model to identify 4G DTU signals, which has the advantages of accuracy, stability, and location adaptability.

[0005] The 4G DTU signal strength identification method based on edge data transmission according to an embodiment of the present invention includes: Collect edge operation data from 4G DTU devices and read the location fingerprint database. After preprocessing, generate edge transmission status sequence and location status data. Service flow features and power consumption features are extracted from the edge transmission state sequence, and cross-layer coupling coding is performed to generate a service-RF coupling feature vector. The edge transmission state sequence and the service-RF coupling feature vector are input into the edge construction layer of the edge LTC signal identification model to generate the edge signal state vector. The edge signal state vector is input into the dual constant layer to generate the signal fluctuation time constant and the transmission anomaly time constant. After mutual modulation, the modulated signal fluctuation time constant and the modulated transmission anomaly time constant are generated. The edge signal state vector, the modulated signal fluctuation time constant, and the modulated transmission anomaly time constant are input into the gated recursive layer to generate transmission gate coefficients. Based on the transmission gate coefficients, the edge signal state vector is gated recursively to generate a signal strength state representation. Based on the modulation signal fluctuation time constant, the modulation transmission abnormality time constant, power consumption characteristics, and point status data, an interlock protection determination is performed, and an RF protective derating instruction is generated. The signal strength status is input to the hysteresis output layer. Based on the signal strength status and the radio frequency protective derating command, the indication control parameters are generated. The display status is output according to the indication control parameters. The location fingerprint database is updated using the calibrated signal strength level, edge transmission status sequence and location status data to generate 4G DTU signal strength identification results.

[0006] Optionally, the edge operation data includes communication status data, edge transmission result data, and DTU real-time power consumption curve; the point fingerprint database includes point reference data corresponding to the current installation location; and the preprocessing includes sampling time correction, invalid record removal, numerical normalization, and time alignment.

[0007] Optionally, the generation of the service-RF coupling feature vector includes: Extract transmission records from the edge transmission state sequence, and obtain service flow characteristics based on the transmission records; Power consumption change records are extracted from the edge transmission state sequence, and power consumption characteristics are obtained based on the power consumption change records. Wireless link layer features are extracted from the edge transmission state sequence, and service perturbation modulation and power offset correction are applied to the wireless link layer features based on service flow features and power consumption features to generate compensated link features. The service flow characteristics, power consumption characteristics, and compensated link characteristics are combined to generate a service-RF coupling feature vector.

[0008] Optionally, the generation of the edge signal state vector includes: The edge transmission state sequence and service-RF coupling feature vector are input into the edge construction layer of the edge LTC signal identification model; The edge LTC signal recognition model includes an edge construction layer, a dual constant layer, a gated recursive layer, and a hysteresis output layer; In the edge construction layer, the edge transmission state sequence is temporally encoded to generate an edge temporal state representation; Scale matching is performed on the service-RF coupling feature vector according to the edge temporal state representation to generate a service-RF matching representation; Establish the correspondence between edge timing state representation and service-RF matching representation, and perform state fusion based on the correspondence to generate edge signal state vector.

[0009] Optionally, the generation of the modulated signal fluctuation time constant and the modulated transmission anomaly time constant includes: The edge signal state vector is input into the dual-constant layer, which includes a signal fluctuation constant generation unit, a transmission anomaly constant generation unit, and a dual-constant intermodulation unit. In the fluctuation constant generation unit, the signal change component and link state component in the edge signal state vector are read, and the signal fluctuation time constant is generated according to the signal change amplitude and link state change trend within the continuous upload cycle. The edge signal state vector is input into the transmission anomaly constant generation unit. The transmission change component and power consumption change component in the edge signal state vector are read, and the transmission anomaly time constant is generated according to the duration of the transmission anomaly and the degree of power consumption deviation within the continuous upload cycle. The dual-constant intermodulation unit takes the signal fluctuation time constant and the transmission anomaly time constant as inputs, performs fading enhancement modulation on the transmission anomaly time constant based on the signal fluctuation time constant, and performs anomaly feedback modulation on the signal fluctuation time constant based on the transmission anomaly time constant, to generate the modulated signal fluctuation time constant and the modulated transmission anomaly time constant.

[0010] Optionally, the generation of the signal strength state representation includes: Input the edge signal state vector, the modulated signal fluctuation time constant, and the modulated transmission anomaly time constant into the gated recursive layer; In the gated recursive layer, a recursive input sequence is generated based on the edge signal state vector, the modulation signal fluctuation time constant, and the modulation transmission anomaly time constant. An abnormal absorption coefficient is generated based on the abnormal transmission time constant after modulation, a stability holding coefficient is generated based on the fluctuation time constant of the modulated signal, and a transmission gating coefficient is generated based on the abnormal absorption coefficient and the stability holding coefficient. Candidate signal states are generated based on the recursive input sequence, and the candidate signal states and the recursive states of the previous upload cycle are updated by gating using transmission gating coefficients to generate the recursive states of the current upload cycle. Repeatedly perform gating updates, and fuse the recursive state of the last upload cycle with the recursive state changes in consecutive upload cycles to generate a signal strength state representation.

[0011] Optionally, the generation of the radio frequency protective derating command includes: Generate signal fading judgment threshold, link degradation judgment threshold, historical power consumption baseline and power consumption protection ratio based on the location status data; The modulation signal fluctuation time constant is compared with the signal fading judgment threshold to generate a signal fading judgment identifier; The abnormal transmission time constant after modulation is compared with the link degradation judgment threshold to generate a link degradation judgment identifier; Generate real-time power deviation indicators based on power consumption characteristics and historical power consumption baseline; Based on the signal fading judgment flag, link degradation judgment flag, and real-time power consumption deviation flag, an RF protective derating command is generated.

[0012] Optionally, the generation of the 4G DTU signal strength identification result includes: The signal strength state representation is input to the hysteresis output layer, an initial signal strength level is generated based on the signal strength state representation, and a dynamic hysteresis threshold is generated according to the signal strength state representation and the radio frequency protective derating instruction. The calibrated signal strength level is generated based on the initial signal strength level, dynamic hysteresis threshold, and radio frequency protective derating command. The signal recognition reliability is then generated based on the calibrated signal strength level and the signal strength state representation. Indication control parameters are generated based on the calibrated signal strength level, signal identification reliability, and radio frequency protection derating instructions; The system outputs and displays the status according to the indicated control parameters, and associates the calibrated signal strength level, edge transmission status sequence, and location status data into the location fingerprint database. It then updates the signal attenuation fingerprint vector in the location fingerprint database and generates the 4GDTU signal strength identification result.

[0013] The beneficial effects of this invention are: The proposed 4G DTU signal strength identification method based on edge data transmission employs an edge LTC signal identification model, forming a continuous identification link through an edge construction layer, a dual-constant layer, a gated recursive layer, and a hysteresis output layer. The dual-constant layer generates signal fluctuation time constants and transmission anomaly time constants, and performs mutual modulation, enabling the model to simultaneously express rapid signal fading and transmission anomaly accumulation. The gated recursive layer generates transmission gate coefficients based on the modulated dual time constants, recursively applying them to the edge signal state vector, improving the model's ability to represent the signal evolution state during continuous upload cycles. The hysteresis output layer further combines RF protective derating instructions to generate indication control parameters, making the final display state more stable and reducing frequent color jumps caused by short-term jitter.

[0014] This invention also enhances field adaptability through interlock protection judgment and location fingerprint database update mechanisms. When the modulated signal fluctuation time constant, the modulated transmission anomaly time constant, power consumption characteristics, and location status data all meet the protection conditions, a radio frequency protective derating command is generated. This command can limit the maximum transmit power level when rapid signal fading, link degradation, and abnormal power consumption occur simultaneously, reducing the risk of power backlash degradation. Simultaneously, the system updates the location fingerprint database using calibrated signal strength levels, edge transmission state sequences, and location status data. This allows subsequent identification boundaries and display strategies to gradually adapt to the actual communication environment of the installation location, improving the reliability of 4G DTU field installation, commissioning, and long-term operation and maintenance. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of the 4G DTU signal strength identification method based on edge data transmission proposed in this invention; Figure 2 This is a schematic diagram of the hierarchical processing and dual-constant intermodulation mechanism of the edge LTC signal identification model of the 4G DTU signal strength identification method based on edge data transmission proposed in this invention. Figure 3 This is a flowchart illustrating the interlock protection determination, hysteresis output control, and point fingerprint database closed-loop update process of the 4G DTU signal strength identification method based on edge data transmission proposed in this invention. Detailed Implementation

[0016] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0017] refer to Figures 1-3 A 4G DTU signal strength identification method based on edge data transmission includes: Collect edge operation data from 4G DTU devices and read the location fingerprint database. After preprocessing, generate edge transmission status sequence and location status data. Service flow features and power consumption features are extracted from the edge transmission state sequence, and cross-layer coupling coding is performed to generate a service-RF coupling feature vector. The edge transmission state sequence and the service-RF coupling feature vector are input into the edge construction layer of the edge LTC signal identification model to generate the edge signal state vector. The edge signal state vector is input into the dual constant layer to generate the signal fluctuation time constant and the transmission anomaly time constant. After mutual modulation, the modulated signal fluctuation time constant and the modulated transmission anomaly time constant are generated. The edge signal state vector, the modulated signal fluctuation time constant, and the modulated transmission anomaly time constant are input into the gated recursive layer to generate transmission gate coefficients. Based on the transmission gate coefficients, the edge signal state vector is gated recursively to generate a signal strength state representation. Based on the modulation signal fluctuation time constant, the modulation transmission abnormality time constant, power consumption characteristics, and point status data, an interlock protection determination is performed, and an RF protective derating instruction is generated. The signal strength status is input to the hysteresis output layer. Based on the signal strength status and the radio frequency protective derating command, the indication control parameters are generated. The display status is output according to the indication control parameters. The location fingerprint database is updated using the calibrated signal strength level, edge transmission status sequence and location status data to generate 4G DTU signal strength identification results.

[0018] In this embodiment, the edge operation data includes communication status data, edge transmission result data, and DTU real-time power consumption curve. The communication status data includes CSQ signal strength value, network registration status, and wireless link layer status. The edge transmission result data includes data packet arrival time, payload size, planned upload time, actual reporting time, upload success flag, upload failure flag, and retransmission count. The DTU real-time power consumption curve includes RF power amplifier current and overall power consumption. The location fingerprint database includes location reference data corresponding to the current installation location. The location reference data includes location identifier, geographic coordinates, historical power consumption baseline, environmental attenuation coefficient, historical signal level record, and historical signal attenuation fingerprint vector. Preprocessing includes sampling time correction, invalid record removal, numerical normalization, and time alignment.

[0019] In this embodiment, the generation of the service-RF coupling feature vector includes: Extract transmission records from the edge transmission state sequence, and obtain service flow characteristics based on the transmission records; The generation of service flow characteristics specifically includes: reading transmission records from the edge transmission state sequence, extracting the arrival time, payload size, and actual reporting time of data packets within the same upload cycle, taking the interval between the arrival times of adjacent data packets as the data packet arrival interval, taking the discrete change in payload size within the same upload cycle as the payload size change, taking the deviation of the actual reporting time from the planned upload time as the edge upload cycle offset, generating the data packet arrival interval entropy based on the data packet arrival interval (the entropy is obtained by statistically analyzing the occurrence ratio of different intervals within the same upload cycle, summing the products of each occurrence ratio and its logarithm, and taking the opposite number), generating the payload size variance within a unit cycle based on the payload size change (the variance is obtained by calculating the average of the squared differences between each payload size and the average payload size within the same upload cycle), generating the edge upload cycle jitter based on the edge upload cycle offset (the jitter is obtained by calculating the average absolute change of the offset of consecutive upload cycles), and normalizing and combining the data packet arrival interval entropy, the payload size variance within a unit cycle, and the edge upload cycle jitter to generate service flow characteristics. The normalization combination of business flow features specifically includes: using the historical minimum and maximum values ​​of the corresponding business type in the current location fingerprint database as normalization boundaries, normalization processing is performed on the data packet arrival interval entropy, the payload size variance within a unit period, and the edge upload period jitter. The normalized data packet arrival interval entropy is obtained by subtracting the historical minimum data packet arrival interval entropy from the current data packet arrival interval entropy, and then dividing by the difference between the historical maximum data packet arrival interval entropy and the historical minimum data packet arrival interval entropy. The normalized payload size variance within a unit period is obtained by subtracting the historical minimum payload size variance from the current payload size variance within a unit period. After calculating the variance of the payload size, divide it by the difference between the historical maximum payload size variance and the historical minimum payload size variance. The normalized edge upload cycle jitter is obtained by subtracting the historical minimum edge upload cycle jitter from the current edge upload cycle jitter, and then dividing it by the difference between the historical maximum edge upload cycle jitter and the historical minimum edge upload cycle jitter. When the historical maximum value is equal to the historical minimum value, the corresponding normalization result is set to 0. The normalized data packet arrival interval entropy, the normalized payload size variance within a unit period, and the normalized edge upload cycle jitter are arranged in order to generate business flow characteristics. Power consumption change records are extracted from the edge transmission state sequence, and power consumption characteristics are obtained based on the power consumption change records. The generation of power consumption characteristics specifically includes: reading power consumption change records from the edge transmission state sequence. These records are determined by the DTU's real-time power consumption curve, which includes the RF power amplifier current and the overall power consumption. The RF power amplifier current change is generated based on the RF power amplifier current, calculated by comparing the difference between the average RF power amplifier current in the current upload cycle and the average RF power amplifier current in the previous upload cycle. Finally, the overall power consumption change is generated based on the overall power consumption, calculated by comparing the average overall power consumption in the current upload cycle with the historical data corresponding to the same upload cycle type in the location fingerprint database. The difference between historical power consumption baselines is obtained. The temperature rise drift is generated based on the change in RF power amplifier current and the change in overall power consumption. The temperature rise drift is obtained by performing normalized weighted summation on the change in RF power amplifier current and the change in overall power consumption. The change in RF power amplifier current, the change in overall power consumption, and the temperature rise drift are taken as three components. Normalization is performed on each component separately. The normalization boundary is determined by the historical minimum and historical maximum values ​​of the installation point in the point fingerprint database. The normalized change in RF power amplifier current, the normalized change in overall power consumption, and the normalized temperature rise drift are arranged in order to generate power consumption characteristics. Wireless link layer features are extracted from the edge transmission state sequence, and service perturbation modulation and power offset correction are applied to the wireless link layer features based on service flow features and power consumption features to generate compensated link features. The generation of compensated link features specifically includes: reading the packet arrival interval entropy, payload size variance within a unit period, and edge upload period jitter from the service flow features, performing normalization and summation on the three to generate a burst index, using the burst index as the service weight, reading the wireless link layer features from the edge transmission state sequence, using the service weight to perform weighted fusion of the wireless link layer features to generate service-weighted link features, reading the temperature rise drift from the power consumption features, using the temperature rise drift as the power consumption offset, performing offset compensation on the service-weighted link features to generate compensated link features; The generation of wireless link layer features specifically includes: reading CSQ signal strength status, network registration status, wireless link layer status and retransmission count from the edge transmission state sequence; arranging the mean, maximum change amplitude and last sampled value in the CSQ signal strength status in order; and concatenating the network registration status, wireless link layer status and retransmission count after converting them into numerical codes to generate wireless link layer features. The service flow characteristics, power consumption characteristics, and compensated link characteristics are combined to generate a service-RF coupling feature vector; The generation of the service-RF coupling feature vector specifically includes: concatenating the service flow features, power consumption features, and compensated link features according to the same edge data upload cycle to generate a concatenated feature vector; reading the historical mean and historical standard deviation of each feature component corresponding to the current installation point from the point fingerprint database; subtracting the corresponding historical mean from each component in the concatenated feature vector and then dividing by the corresponding historical standard deviation to obtain the standardized component; when the historical standard deviation is 0, setting the corresponding standardized component to 0; arranging all standardized components according to the component order before concatenation to generate the service-RF coupling feature vector.

[0020] In this embodiment, the generation of the edge signal state vector includes: The edge transmission state sequence and service-RF coupling feature vector are input into the edge construction layer of the edge LTC signal identification model; The edge LTC signal recognition model includes an edge construction layer, a dual constant layer, a gated recursive layer, and a hysteresis output layer; The training process of the edge LTC signal recognition model includes: constructing a training sample set, which includes sample edge transmission state sequences, sample service-RF coupling feature vectors, sample RF protective derating instructions, sample signal strength level labels, sample signal recognition confidence labels, and sample display status labels. The sample edge transmission state sequences are derived from historical 4G... The sample service-RF coupling feature vector is obtained through DTU edge operation data preprocessing. It is encoded by cross-layer coupling of service flow features and power consumption features in the sample edge transmission state sequence. The sample signal strength level label is determined by the measured CSQ signal strength value, historical successful upload records, and manually verified communication quality at the installation point. The sample signal recognition reliability label is determined by the consistency between the sample signal strength level label and the actual edge upload result. The sample display status label is determined by historical display color, breathing cycle, and protection prompt status records. During training, the sample edge transmission state sequence and sample service-RF coupling feature vector are input into the edge construction layer of the edge LTC signal recognition model to generate the sample edge signal state vector. This vector is then input into a dual-constant layer to generate the sample modulated signal fluctuation time constant and the sample modulated transmission anomaly time constant. These are then input into a gated recursive layer to generate the sample signal state vector. The signal strength state representation and the sample RF protective derating command are input into the hysteresis output layer to generate the sample calibrated signal strength level, sample signal recognition confidence, and sample indication control parameters. The loss function includes signal level loss, confidence loss, display state loss, and hysteresis stability loss. The total loss is obtained by adding the above losses. The signal level loss is used to constrain the sample calibrated signal strength level to be consistent with the sample signal strength level label. The confidence loss is used to constrain the sample signal recognition confidence to be consistent with the sample signal recognition confidence label. The display state loss is used to constrain the display state corresponding to the sample indication control parameters to be consistent with the sample display state label. The hysteresis stability loss is used to constrain the output level of adjacent upload cycles to remain stable when the dynamic hysteresis threshold is not reached. An adaptive moment estimation optimizer is used to update the trainable parameters in the edge construction layer, the dual constant layer, the gated recursive layer, and the hysteresis output layer. The training convergence condition is that the total loss decrease is less than the convergence value in multiple consecutive training rounds. After training, an edge LTC signal recognition model for deployment at the edge of 4G DTU equipment is obtained. Hysteresis stability loss is used to constrain the level output stability of the hysteresis output layer. Its construction process specifically includes: reading the initial signal strength level of the sample, the signal strength level after sample calibration, the dynamic hysteresis threshold of the sample, and the signal strength level of the sample in the previous display period. When the level difference between the initial signal strength level of the sample and the signal strength level in the previous display period of the sample does not reach the dynamic hysteresis threshold of the sample, the signal strength level after sample calibration is constrained to remain at the signal strength level of the previous display period of the sample. When the level difference reaches the dynamic hysteresis threshold of the sample, the signal strength level after sample calibration is constrained to update with the initial signal strength level of the sample. The level difference generated by the above two types of constraints is averaged to generate hysteresis stability loss. The improvements to the edge LTC signal identification model include: 1. Compared to the original LTC model, which generates a single liquid time constant based solely on the input state, the edge LTC signal identification model of this invention sets up a dual-constant layer to generate signal fluctuation time constants and transmission anomaly time constants respectively. Mutual modulation is performed within the dual-constant layer, enabling rapid signal fading to enhance the transmission anomaly time constant and allowing transmission anomalies to provide feedback correction to the signal fluctuation time constant. This solves the problem of the original LTC model's difficulty in distinguishing between the natural fluctuations of wireless signals and the coupling effects of edge uplink anomalies. 2. Compared to the original LTC model, which only updates according to the hidden state, the edge LTC signal identification model of this invention sets up a gated recursion layer. The modulated signal fluctuation time constant and the modulated transmission anomaly time constant are used together to generate transmission gating coefficients. These transmission gating coefficients control the contribution of the current uplink cycle state and the previous uplink cycle recursion state to the current recursion state in the form of absorption and retention ratios, respectively. This allows the model to enhance the absorption of the current anomaly state when the link is abnormal and enhance the retention of historical states when the signal is stable, thereby improving 4G performance. The DTU signal strength identification model has the following characteristics: First, it has the ability to respond to sudden retransmissions, upload delays and increased power consumption. Second, compared with the original LTC model which directly outputs classification results, the edge LTC signal identification model of this invention sets up a hysteresis output layer. Based on the signal strength state representation and RF protection derating instructions, it generates calibrated signal strength levels, signal identification reliability and indication control parameters. This allows the model output to simultaneously adapt to RGB display stability and RF protection status, avoids frequent jumps in local indicator units caused by short-term signal jitter, and enables the identification results to be used for subsequent point fingerprint database updates. In the edge construction layer, the edge transmission state sequence is temporally encoded to generate an edge temporal state representation; The generation of edge temporal state representation specifically includes: arranging periodic states according to the chronological order of edge data upload cycles, where a periodic state is a state record formed in the edge transmission state sequence with one edge data upload cycle as a unit, including the communication state, transmission state, and power consumption state within the upload cycle; numerically encoding the discrete states in the periodic states; scaling the continuous states in the periodic states so that the communication state, transmission state, and power consumption state can be represented side by side in the same periodic state; arranging the scale-scaled periodic states in temporal order, with the state code corresponding to each upload cycle as a row, and arranging the state codes corresponding to multiple consecutive upload cycles in chronological order to generate the edge temporal state representation; The specific numerical encoding of discrete states includes: the discrete states in the periodic state include network registration state, wireless link layer state, and upload result. The network registration state is encoded as 0, 1, and 2 respectively for unregistered, registering, and registered. The wireless link layer state is encoded as 0, 1, and 2 respectively for link disconnected, link unstable, and link stable. The upload result is encoded as 0, 1, and 2 respectively for upload failure, retransmission success, and first upload success. If there are multiple upload results in the same upload period, the upload result encoding for that upload period is determined in the order of upload failure first, retransmission success second, and first upload success last. The scaling of continuous states specifically includes: continuous states in periodic states include CSQ signal strength, retransmission count, upload delay, RF power amplifier current, and overall power consumption. The upload delay is obtained by subtracting the planned upload time from the actual reporting time. Scaling is achieved using the minimum-maximum normalization method. Specifically, the historical minimum and maximum values ​​of the corresponding state components in the location fingerprint database are read. The current state component is subtracted from the historical minimum value and then divided by the difference between the historical maximum and the historical minimum value to obtain the scaled state component. When the historical maximum value is equal to the historical minimum value, the scaling result of the state component is set to 0. The generation of state codes specifically includes: arranging the discrete states that have completed numerical coding and the continuous states that have completed the coding at a unified scale within the same upload cycle in the order of CSQ signal strength value, network registration status, wireless link layer status, upload result, number of retransmissions, upload delay, RF power amplifier current, and total power consumption to generate the state code corresponding to that upload cycle. The state code is a set of numerical sequences arranged in a fixed order. Scale matching is performed on the service-RF coupling feature vector according to the edge temporal state representation to generate a service-RF matching representation; The generation of the service-RF matching representation specifically includes: reading the service burst component, wireless link change component, and power consumption change component from the service-RF coupling feature vector; performing period matching on the service-RF coupling feature vector based on the number of upload cycles in the edge timing state representation; when the number of cycles corresponding to the service-RF coupling feature vector is less than the number of cycles corresponding to the edge timing state representation, padding is performed according to the adjacent cycle continuation method; when the number of cycles corresponding to the service-RF coupling feature vector is greater than the number of cycles corresponding to the edge timing state representation, merging is performed according to the same upload cycle merging method; and performing scale matching on the period-matched service-RF coupling feature vector based on the feature scale in the edge timing state representation. Scale matching includes component normalization and dimension alignment. After completing period matching and scale matching, the service-RF matching representation is generated. The adjacent cycle continuation method specifically includes: for an upload cycle that lacks a service-RF coupling feature vector in the edge timing state representation, find the most recent upload cycle that has formed a service-RF coupling feature vector before this upload cycle, copy the service-RF coupling feature vector of the most recent upload cycle to the upload cycle that lacks data; if there is no upload cycle that has formed a service-RF coupling feature vector before this upload cycle, find the most recent upload cycle that has formed a service-RF coupling feature vector after this upload cycle, and copy its service-RF coupling feature vector. The specific method for merging features with the same upload cycle includes: grouping multiple service-RF coupling feature vectors with the same upload cycle start time and upload cycle end time into the same cycle group; merging the service-RF coupling feature vectors within the same cycle group according to their component positions; for component positions corresponding to service flow features and power consumption features, using the average value of the corresponding component values ​​within the same cycle group as the merge value; for component positions corresponding to compensated link features, if there is a service-RF coupling feature vector corresponding to an upload failure within the same cycle group, then using the component value corresponding to the compensated link feature in the service-RF coupling feature vector corresponding to the upload failure as the merge value. The radio frequency coupling feature vector is the service-radio frequency coupling feature vector generated when the upload result is an upload failure. If there are multiple service-radio frequency coupling feature vectors corresponding to upload failures within the same period group, the maximum value of the same compensated link feature component position among the multiple service-radio frequency coupling feature vectors corresponding to upload failures is used as the merging value. If there are no service-radio frequency coupling feature vectors corresponding to upload failures within the same period group, the average value of the component values ​​corresponding to the compensated link feature within the same period group is used as the merging value. The merging values ​​corresponding to the service flow feature, power consumption feature, and compensated link feature are arranged in order to generate a service-radio frequency coupling feature vector corresponding to the upload period. Component normalization specifically includes: reading each component in the service-RF coupling feature vector after period matching, performing minimum-maximum normalization according to the historical minimum and historical maximum values ​​of the corresponding component in the location fingerprint database, so that each component in the service-RF coupling feature vector is in the same numerical range as the edge timing state representation, and when the historical maximum value is equal to the historical minimum value, the normalization result of the corresponding component is set to 0; Dimensional alignment specifically includes: determining the number of state-coded components in each row of the edge temporal state representation, and determining the number of components in the normalized service-radio frequency coupling feature vector. If the number of components in the service-radio frequency coupling feature vector is less than the number of state-coded components, the corresponding components are cyclically added in the order of service burst components, radio link change components, and power consumption change components until the number of components is consistent. If the number of components in the service-radio frequency coupling feature vector is more than the number of state-coded components, the adjacent components that belong to the same type are averaged and merged until the number of components is consistent. Establish the correspondence between edge temporal state representation and service-RF matching representation, and perform state fusion based on the correspondence to generate edge signal state vector; The generation of the periodic fusion state specifically includes: establishing a correspondence between the edge temporal state representation and the service-radio frequency matching representation according to the same edge data upload cycle. Specifically, the start time and end time of the upload cycle are used as the matching key. A certain row of state codes in the edge temporal state representation is bound to a row of service-radio frequency matching codes with the same upload cycle identifier in the service-radio frequency matching representation to form a one-to-one correspondence. When the upload cycle identifiers cannot be completely consistent, the service-radio frequency matching code with the largest time overlap ratio is selected to correspond to the state code of that row. The time overlap ratio is obtained by dividing the overlap duration of the two upload cycles by the upload cycle duration corresponding to the state code of that row. The state codes and service-radio frequency matching codes corresponding to the same upload cycle are concatenated side by side, with the state code placed first and the service-radio frequency matching code placed last to form the periodic fusion state of that upload cycle. The generation of the edge signal state vector specifically includes: reading the periodic fusion state according to the order of edge data upload cycles, performing state compression on multiple consecutive periodic fusion states, wherein state compression includes averaging the components at the same position in multiple consecutive upload cycles to obtain the average state component, calculating the maximum change amplitude of the components at the same position in multiple consecutive upload cycles to obtain the changed state component, and sequentially concatenating the average state component and the changed state component to generate the edge signal state vector.

[0021] In this embodiment, the generation of the modulated signal fluctuation time constant and the modulated transmission anomaly time constant includes: The edge signal state vector is input into the dual-constant layer, which includes a signal fluctuation constant generation unit, a transmission anomaly constant generation unit, and a dual-constant intermodulation unit. In the fluctuation constant generation unit, the signal change component and link state component in the edge signal state vector are read, and the signal fluctuation time constant is generated according to the signal change amplitude and link state change trend within the continuous upload cycle. The generation of the signal fluctuation time constant specifically includes: reading the signal change component and link state component from the edge signal state vector. The signal change component is determined by the component in the edge signal state vector corresponding to the CSQ signal strength change, and the link state component is determined by the component in the edge signal state vector corresponding to the network registration state and the wireless link layer state. The change amplitude of the CSQ signal strength value within a continuous upload cycle is calculated. The absolute value of the difference between the CSQ signal strength values ​​in adjacent upload cycles is taken as the single-cycle signal change, and the average value of multiple consecutive single-cycle signal changes is taken as the signal fluctuation. The number of changes in the link state code within a continuous upload cycle is calculated. The number of changes in the network registration state code and the number of changes in the wireless link layer state code are added together to generate the link state change. The signal fluctuation and the link state change are normalized respectively. The results are weighted and summed to generate the signal fluctuation time constant. The normalization process uses the historical minimum and historical maximum values ​​of the same type of state component in the point fingerprint database as boundaries. When the historical maximum value is equal to the historical minimum value, the corresponding normalization result is set to 0. The edge signal state vector is input into the transmission anomaly constant generation unit. The transmission change component and power consumption change component in the edge signal state vector are read, and the transmission anomaly time constant is generated according to the duration of the transmission anomaly and the degree of power consumption deviation within the continuous upload cycle. The generation of the transmission anomaly time constant specifically includes: reading the transmission change component and power consumption change component from the edge signal state vector. The transmission change component is determined by the components in the edge signal state vector corresponding to the upload result, retransmission count, and upload delay. The power consumption change component is determined by the components in the edge signal state vector corresponding to the RF power amplifier current and overall power consumption. The number of abnormal upload cycles is calculated. The number of abnormal upload cycles equals the number of cycles corresponding to upload failures within a continuous upload cycle plus the number of cycles corresponding to successful retransmissions. The number of abnormal upload cycles is divided by the total number of continuous upload cycles to generate the upload anomaly duration. The continuous upload duration is then calculated. The difference between the average upload delay within a period and the historical upload delay baseline is used to generate the upload delay deviation. The deviation of the RF power amplifier current and the total power consumption relative to the historical power consumption baseline within a continuous upload period is calculated to generate the power consumption deviation. The upload anomaly duration, upload delay deviation, and power consumption deviation are normalized to obtain the normalized results. The normalized upload anomaly duration and normalized upload delay deviation are weighted and summed to generate the transmission anomaly base quantity. The normalized power consumption deviation is added to 1 to obtain the power consumption correction factor. The transmission anomaly base quantity is multiplied by the power consumption correction factor to generate the transmission anomaly time constant. The dual-constant intermodulation unit takes the signal fluctuation time constant and the transmission anomaly time constant as inputs, performs fading enhancement modulation on the transmission anomaly time constant based on the signal fluctuation time constant, and performs anomaly feedback modulation on the signal fluctuation time constant based on the transmission anomaly time constant, to generate the modulated signal fluctuation time constant and the modulated transmission anomaly time constant. The generation of the modulation-based transmission anomaly time constant specifically includes: reading the signal fluctuation time constant and the environmental attenuation coefficient in the point status data, generating the transmission anomaly modulation factor, which is equal to 1 plus the product of the environmental attenuation coefficient and the signal fluctuation time constant; multiplying the transmission anomaly modulation factor by the transmission anomaly time constant to generate the modulation-based transmission anomaly time constant; the modulation-based transmission anomaly time constant can better reflect the enhanced influence of rapid signal fading on the transmission anomaly state. The generation of the modulation signal fluctuation time constant specifically includes: reading the transmission anomaly time constant and the transmission change component in the edge signal state vector, generating a signal fluctuation feedback factor. The signal fluctuation feedback factor is equal to 1 plus the product of the transmission feedback coefficient corresponding to the transmission change component and the transmission anomaly time constant. The signal fluctuation feedback factor is multiplied by the signal fluctuation time constant to generate the modulation signal fluctuation time constant. The modulation signal fluctuation time constant can better reflect the feedback influence of transmission anomaly on signal fluctuation judgment. The generation of the environmental attenuation coefficient specifically includes: reading the historical signal level record, geographical coordinates, historical power consumption baseline, and historical signal attenuation fingerprint vector corresponding to the current installation point from the point status data; calculating the point signal attenuation based on the historical signal level record, which is obtained by averaging the absolute values ​​of the differences in historical signal levels between adjacent upload cycles; calculating the point attenuation stability based on the historical signal attenuation fingerprint vector, which is obtained by averaging the absolute values ​​of the differences between each component in the historical signal attenuation fingerprint vector and its average component; and calculating the point power consumption based on the historical power consumption baseline and the current power consumption status. The deviation is obtained by subtracting the historical power consumption baseline from the current average power consumption of the whole machine, and then dividing by the historical power consumption baseline. When the historical power consumption baseline is 0, the average non-zero historical power consumption baseline of the same installation environment in the point fingerprint library is used as the substitute. The point signal attenuation, point attenuation stability and point power consumption deviation are normalized respectively. The normalization boundary is the historical minimum and historical maximum values ​​of the corresponding components of the same installation environment in the point fingerprint library. The average of the normalized point signal attenuation, normalized point attenuation stability and normalized point power consumption deviation is calculated to generate the environmental attenuation coefficient. The generation of the transmission feedback coefficient specifically includes: reading the upload result corresponding component, retransmission number corresponding component, and upload delay corresponding component from the transmission change components in the edge signal state vector; generating an upload result feedback quantity based on the upload result corresponding component; setting the upload result feedback quantity to 0 when the upload result corresponding component indicates a successful first upload; setting the upload result feedback quantity to half when the upload result corresponding component indicates a successful retransmission; and setting the upload result feedback quantity to 1 when the upload result corresponding component indicates an upload failure. The retransmission feedback quantity is generated based on the retransmission number corresponding component. The retransmission feedback quantity is obtained by subtracting the average historical retransmission number in the location fingerprint database from the current retransmission number, and then dividing by the difference between the maximum historical retransmission number and the average historical retransmission number. When the difference is 0, the retransmission feedback is generated. The quantity is set to 0. A delay feedback quantity is generated based on the corresponding component of the upload delay. The delay feedback quantity is obtained by subtracting the average historical upload delay in the location fingerprint database from the current upload delay, and then dividing by the difference between the maximum historical upload delay and the average historical upload delay. When the difference is 0, the delay feedback quantity is set to 0. Smoothing compression mapping is performed on the upload result feedback quantity, retransmission feedback quantity, and delay feedback quantity so that the values ​​of the three are all between 0 and 1. The average of the amplitude-limited upload result feedback quantity, amplitude-limited retransmission feedback quantity, and amplitude-limited delay feedback quantity is calculated to generate the transmission feedback coefficient. Specifically, smooth compression is to divide the feedback quantity by 1 and the sum of the absolute value of the feedback quantity to obtain the compressed feedback quantity. When the feedback quantity is less than 0, the compressed feedback quantity is set to 0. When the feedback quantity is greater than or equal to 0, the compressed feedback quantity is retained.

[0022] In this embodiment, the generation of the signal strength state representation includes: Input the edge signal state vector, the modulated signal fluctuation time constant, and the modulated transmission anomaly time constant into the gated recursive layer; In the gated recursive layer, a recursive input sequence is generated based on the edge signal state vector, the modulation signal fluctuation time constant, and the modulation transmission anomaly time constant. The generation of the recursive input sequence specifically includes: reading the periodic state components corresponding to consecutive upload cycles in the edge signal state vector. Each periodic state component includes communication state, transmission state, power consumption state, and service-RF coupling change. Reading the modulation signal fluctuation time constant and the modulation transmission anomaly time constant, and writing the modulation signal fluctuation time constant and the modulation transmission anomaly time constant to the end of the periodic state component of the corresponding upload cycle according to the cycle identifier of the edge data upload cycle, generating periodic recursive input, arranging multiple periodic recursive inputs according to the order of edge data upload cycles, and generating a recursive input sequence. An abnormal absorption coefficient is generated based on the abnormal transmission time constant after modulation, a stability holding coefficient is generated based on the fluctuation time constant of the modulated signal, and a transmission gating coefficient is generated based on the abnormal absorption coefficient and the stability holding coefficient. The generation of the transmission gating coefficient specifically includes: reading the modulated transmission anomaly time constant corresponding to the current upload cycle, performing smooth compression on the modulated transmission anomaly time constant to generate anomaly absorption coefficient. Specifically, smooth compression involves dividing the modulated transmission anomaly time constant by 1 and summing the absolute value of the modulated transmission anomaly time constant to obtain the anomaly absorption coefficient. Then, the modulated signal fluctuation time constant corresponding to the current upload cycle is read, and a stability holding coefficient is generated based on this time constant. The stability holding coefficient is equal to 1 minus the smooth compression result of the modulated signal fluctuation time constant. The smooth compression result is the modulated signal fluctuation... The value obtained by dividing the time constant by 1 and summing it with the absolute value of the time constant of the modulated signal fluctuation is used to normalize and synthesize the abnormal absorption coefficient and the stability retention coefficient to generate the transmission gate coefficient. The normalization synthesis specifically includes dividing the two coefficients by the sum of the two coefficients to obtain two ratios. The ratio with the abnormal absorption coefficient as the numerator is the absorption ratio of the current cycle state, and the ratio with the stability retention coefficient as the numerator is the retention ratio of the previous uplink cycle recursive state. When the sum of the two coefficients is 0, both the absorption ratio and the retention ratio are set to one-half. The absorption ratio and the retention ratio are combined to form the transmission gate coefficient, which is a gate binary. Candidate signal states are generated based on the recursive input sequence, and the candidate signal states and the recursive states of the previous upload cycle are updated by gating using transmission gating coefficients to generate the recursive states of the current upload cycle. The generation of candidate signal states specifically includes: reading the periodic recursive input of the current upload cycle, performing state aggregation on the communication state component, transmission state component, power consumption state component, and service-RF coupling change component in the periodic recursive input to generate candidate signal states. Specifically, the state aggregation involves grouping each component in the same periodic recursive input according to its state type, averaging the components within each state type to obtain the communication aggregation component, transmission aggregation component, power consumption aggregation component, and coupling aggregation component, and then arranging them in the order of communication aggregation component, transmission aggregation component, power consumption aggregation component, and coupling aggregation component to generate candidate signal states. The generation of the recursive state of the current upload cycle specifically includes: reading the recursive state of the previous upload cycle and the candidate signal state of the current upload cycle; when the current upload cycle is the first upload cycle of the recursive input sequence, the candidate signal state of the upload cycle is used as the initial recursive state; the candidate signal state and the recursive state of the previous upload cycle are updated by gating using the transmission gating coefficient to generate the recursive state of the current upload cycle; the gating update specifically multiplies each component in the candidate signal state by the absorption ratio to generate the current state contribution; multiplies each component in the recursive state of the previous upload cycle by the retention ratio to generate the historical state contribution; and adds the current state contribution and the historical state contribution to generate the recursive state of the current upload cycle. Repeatedly perform gating updates, and fuse the recursive state of the last upload cycle with the recursive state changes in consecutive upload cycles to generate a signal strength state representation; The generation of the signal strength state representation specifically includes: according to the order of edge data upload cycles, repeatedly performing candidate signal state generation, transmission gate coefficient generation, and recursive state generation for each cycle of the recursive input sequence to obtain a recursive state sequence within consecutive upload cycles; reading the recursive state of the last upload cycle in the recursive state sequence and calculating the change between adjacent recursive states within consecutive upload cycles; averaging the components at the same position in all changes to generate an average change vector, wherein the average change vector has the same vector dimension as the recursive state of the last upload cycle; taking the recursive state of the last upload cycle as the first vector segment and the average change vector as the second vector segment; and performing vector concatenation in the order of the first vector segment first and the second vector segment last to generate the signal strength state representation.

[0023] In this embodiment, the generation of the radio frequency protective derating command includes: Generate signal fading judgment threshold, link degradation judgment threshold, historical power consumption baseline and power consumption protection ratio based on the location status data; The generation of the signal fading judgment threshold specifically includes: determining it from the historical signal fluctuation records of the same installation point in the location fingerprint database. Specifically, the signal fluctuation time constant in the historical signal fluctuation record in the normal uploading state is read, its historical mean and historical standard deviation are calculated, and the historical mean is added to twice the historical standard deviation to obtain the signal fading judgment threshold. When the number of historical records for the installation point is insufficient, the historical signal fluctuation records of the same installation environment in the location fingerprint database are read, and the signal fading judgment threshold is obtained in the same way. The generation of the link degradation judgment threshold specifically includes: determining it from the historical transmission anomaly records of the same installation point in the point fingerprint database. Specifically, it involves reading the transmission anomaly time constants in the historical transmission anomaly records that are in a normal uploading state, calculating their historical mean and historical standard deviation, and adding the historical mean to twice the historical standard deviation to obtain the link degradation judgment threshold. When the number of historical transmission anomaly records for the installation point is insufficient, it involves reading the historical transmission anomaly records of the same installation environment in the point fingerprint database and obtaining the link degradation judgment threshold in the same way. The generation of historical power consumption baseline and power consumption protection ratio specifically includes: reading the overall power consumption change component and RF power amplifier current change component in the power consumption characteristics, and reading the historical power consumption baseline and power consumption protection ratio in the point status data. The historical power consumption baseline is the average overall power consumption of the current installation point in the point fingerprint database under normal upload status, and the power consumption protection ratio is 1.5. The modulation signal fluctuation time constant is compared with the signal fading judgment threshold to generate a signal fading judgment identifier; The generation of the signal fading judgment flag specifically includes: comparing the modulation signal fluctuation time constant with the signal fading judgment threshold; when the modulation signal fluctuation time constant is greater than the signal fading judgment threshold, the signal fading judgment flag is set to rapid signal fading; when the modulation signal fluctuation time constant is less than or equal to the signal fading judgment threshold, the signal fading judgment flag is set to no rapid signal fading. The abnormal transmission time constant after modulation is compared with the link degradation judgment threshold to generate a link degradation judgment identifier; The generation of the link degradation judgment flag specifically includes: comparing the abnormal time constant of the modulated transmission with the link degradation judgment threshold; when the abnormal time constant of the modulated transmission is greater than the link degradation judgment threshold, the link degradation judgment flag is set to link degradation; when the abnormal time constant of the modulated transmission is less than or equal to the link degradation judgment threshold, the link degradation judgment flag is set to link not degraded. Generate real-time power deviation indicators based on power consumption characteristics and historical power consumption baseline; The generation of the real-time power consumption deviation flag specifically includes: restoring the current average power consumption of the whole machine based on the power consumption change component in the power consumption characteristics; restoring the current average RF power amplifier current based on the RF power amplifier current change component in the power consumption characteristics; comparing the current average power consumption of the whole machine with the historical power consumption baseline; and comparing the current average RF power amplifier current with the historical RF power amplifier current baseline in the point status data. When the current average power consumption of the whole machine is greater than the product of the historical power consumption baseline and the power consumption protection ratio, and the current average RF power amplifier current is greater than the historical RF power amplifier current baseline, the real-time power consumption deviation flag is set to power consumption exceeding the limit. When the above two conditions are not met at the same time, the real-time power consumption deviation flag is set to power consumption not exceeding the limit. Based on the signal fading judgment flag, link degradation judgment flag, and real-time power consumption deviation flag, an RF protective derating command is generated. The generation of the radio frequency (RF) protective derating instruction specifically includes: reading the signal fading judgment flag, link degradation judgment flag, and real-time power consumption deviation flag. When the signal fading judgment flag indicates rapid signal fading, the link degradation judgment flag indicates link degradation, and the real-time power consumption deviation flag indicates power consumption exceeding the limit, the RF protection conditions are determined to be met. When the RF protection conditions are met, the current transmit power level and the minimum protected transmit power level are read from the point status data. The current transmit power level is reduced by one level and compared with the minimum protected transmit power level. When the transmit power level after reducing by one level is not lower than the minimum protected transmit power level, the transmit power level after reducing by one level is taken as the target transmit power level. When the transmit power level after reducing by one level is lower than the minimum protected transmit power level, the minimum protected transmit power level is taken as the target transmit power level, and an RF protective derating instruction is generated. The RF protective derating instruction includes a protection trigger flag and a target transmit power level. The protection trigger flag indicates that RF protection has been triggered, and the target transmit power level is used to limit 4G. When the maximum transmit power level of the DTU device does not meet the radio frequency protection conditions, the current transmit power level in the point status data is read, the current transmit power level is used as the target transmit power level, and a radio frequency protection derating instruction is generated. At this time, the protection trigger flag in the radio frequency protection derating instruction is used to indicate that radio frequency protection has not been triggered, and the target transmit power level is used to maintain the current transmit power level of the 4G DTU device. The determination of the current transmit power level specifically includes: the transmit power level of the 4G DTU equipment is divided into four levels from low to high: Level 1, Level 2, Level 3, and Level 4. Level 1 represents the lowest transmit power level, with an equivalent maximum transmit power of less than or equal to 10dBm; Level 2 represents the low transmit power level, with an equivalent maximum transmit power of greater than 10dBm and less than or equal to 17dBm; Level 3 represents the medium transmit power level, with an equivalent maximum transmit power of greater than 17dBm and less than or equal to 21dBm; and Level 4 represents the high transmit power level, with an equivalent maximum transmit power of greater than 21dBm and less than or equal to 23dBm. Each transmit power level corresponds to a maximum transmit power control level supported by the 4G communication module. The specific power value is determined by the RF configuration table of the 4G communication module. The maximum transmit power control level in the current RF configuration parameters of the 4G DTU equipment is read, and the maximum transmit power control level is matched with the transmit power level in the RF configuration table to generate the current transmit power level. The generation of the minimum protection transmit power level specifically includes: reading the historical upload success rate, historical signal level record, and historical power consumption baseline from the location status data; when the historical upload success rate is lower than the minimum upload hold success rate in the location fingerprint database, the minimum protection transmit power level is determined to be level three; when the historical upload success rate is not lower than the minimum upload hold success rate and the period of the weak signal level corresponding to the historical signal level record exceeds one-half, the minimum protection transmit power level is determined to be level two; when the historical upload success rate is not lower than the minimum upload hold success rate and the period of the weak signal level corresponding to the historical signal level record does not exceed one-half, the minimum protection transmit power level is determined to be level one. The minimum upload hold success rate is determined by the location fingerprint database based on the normal service upload requirements in the most recent continuous upload period of the installation location. The process of comparing transmission power levels specifically includes: mapping Level 1, Level 2, Level 3, and Level 4 to Level Number 1, Level Number 2, Level Number 3, and Level Number 4, respectively. The larger the level number, the higher the maximum allowable transmission power level. Subtracting one from the level number of the current transmission power level generates a candidate transmission power level number. The candidate transmission power level number is compared with the minimum protected transmission power level number. When the candidate transmission power level number is greater than or equal to the minimum protected transmission power level number, the transmission power level corresponding to the candidate transmission power level number is taken as the target transmission power level. When the candidate transmission power level number is less than the minimum protected transmission power level number, the minimum protected transmission power level is taken as the target transmission power level.

[0024] In this embodiment, the generation of 4G DTU signal strength identification results includes: The signal strength state representation is input to the hysteresis output layer, an initial signal strength level is generated based on the signal strength state representation, and a dynamic hysteresis threshold is generated according to the signal strength state representation and the radio frequency protective derating instruction. The generation of the initial signal strength level specifically includes: reading the component corresponding to the signal strength in the most recent recursive state, comparing the component with the weak signal level boundary, medium signal level boundary, good signal level boundary and excellent signal level boundary of the corresponding installation point in the point fingerprint database, and generating the initial signal strength level. The level boundary is determined by the historical signal level records and historical successful upload records in the point fingerprint database. Reading the component corresponding to the signal strength specifically includes: reading the most recent recursive state in the signal strength state representation, where the most recent recursive state is the recursive state corresponding to the last edge data upload cycle in the recursive state sequence, which is a numerical vector; reading the component corresponding to the CSQ signal strength state from the most recent recursive state as the signal strength identification value, specifically the state component formed by recursing the CSQ signal strength value in the edge transmission state sequence through the edge construction layer, the dual constant layer, and the gated recursive layer, which is used to characterize the signal strength state corresponding to the most recent upload cycle; The generation of grade boundaries specifically includes: reading historical signal grade records and historical successful upload records of the current installation point from the point fingerprint database; assigning the historical signal strength identification values ​​corresponding to weak signal grade, medium signal grade, good signal grade, and excellent signal grade in the historical signal grade records to the corresponding grade sets; reading the upload success identifiers of each historical upload cycle from the historical successful upload records; determining the historical signal strength identification values ​​corresponding to the historical upload cycles with successful upload identifiers as valid historical signal strength identification values; using the valid historical signal strength identification values ​​to filter each grade set, deleting the historical signal strength identification values ​​corresponding to the upload failure cycles; and generating a valid weak signal grade set and a valid medium signal grade set. The effective set of good signal levels and the effective set of excellent signal levels are combined. The maximum historical signal strength identification value in the effective weak signal level set is used as the upper boundary of the weak signal level, the minimum historical signal strength identification value in the effective medium signal level set is used as the lower boundary of the medium signal level, the maximum historical signal strength identification value in the effective medium signal level set is used as the upper boundary of the medium signal level, the minimum historical signal strength identification value in the effective good signal level set is used as the lower boundary of the good signal level, and the maximum historical signal strength identification value in the effective good signal level set is used as the upper boundary of the good signal level, and the minimum historical signal strength identification value in the effective excellent signal level set is used as the lower boundary of the excellent signal level, thus forming the level boundaries. The generation of historical signal level records specifically includes: storing historical signal level records formed by the current installation point within the historical edge data upload cycle in the point fingerprint database, including historical CSQ signal strength value, historical signal strength identification value, historical calibrated signal strength level, and historical upload success identifier. The historical signal strength identification value is obtained by processing the historical CSQ signal strength value through the edge LTC signal identification model. When the installation point has not yet formed enough model history records, the historical CSQ signal strength value is used as the historical signal strength identification value, where the historical signal strength identification value is an integer between 0 and 31. When the historical signal strength identification value is 0 to 17, it is classified into the weak signal level set; when the historical signal strength identification value is 18 to 22, it is classified into the medium signal level set; when the historical signal strength identification value is 23 to 27, it is classified into the good signal level set; and when the historical signal strength identification value is 28 to 31, it is classified into the excellent signal level set. The generation of the dynamic hysteresis threshold specifically includes: reading the component corresponding to the signal strength change in the average change vector to generate the state change amplitude; reading the protection trigger identifier and target transmit power level in the radio frequency protective derating instruction to generate the protection impact quantity. Specifically, when the protection trigger identifier indicates that it has been triggered, the protection impact quantity is obtained by subtracting the target transmit power level number from the current transmit power level number; when the protection trigger identifier indicates that it has not been triggered, the protection impact quantity is 0. The dynamic hysteresis threshold is generated based on the state change amplitude and the protection impact quantity. The dynamic hysteresis threshold is equal to the sum of the basic hysteresis threshold, the threshold increment corresponding to the state change amplitude, and the threshold increment corresponding to the protection impact quantity. The basic hysteresis threshold is determined by the average jump amplitude of the historical signal level jump records of the same installation point in the point fingerprint database. The threshold increment corresponding to the state change amplitude is obtained by multiplying the state change amplitude by the basic hysteresis threshold. The threshold increment corresponding to the protection impact quantity is obtained by multiplying the protection impact quantity by the basic hysteresis threshold. The generation of the basic hysteresis threshold specifically includes: reading historical signal level jump records for the same installation point from the point fingerprint database. These historical signal level jump records include the historical display period, the signal strength level before the jump, the signal strength level after the jump, and the corresponding signal strength identification value at the time of the jump. Weak signal level, medium signal level, good signal level, and excellent signal level are mapped to level number one, level number two, level number three, and level number four, respectively. The level number before and after each historical jump is read, and the absolute value of the difference between the level number after the jump and the level number before the jump is calculated to generate the single level jump amplitude. This is applied to all installation points at the same installation point. The average level jump amplitude is generated by averaging the amplitude of each single level jump. The signal strength identification value corresponding to each historical jump is read, and the absolute value of the difference between the signal strength identification values ​​corresponding to two adjacent historical jumps is calculated to generate the single identification value jump amplitude. The average identification value jump amplitude is generated by averaging all single identification value jump amplitudes. The average level jump amplitude and the average identification value jump amplitude are normalized and averaged to generate the basic hysteresis threshold. When the number of historical signal level jump records at the same installation point is insufficient, the historical signal level jump records of the same installation environment in the point fingerprint database are read, and the basic hysteresis threshold is generated in the same way. The generation of the threshold increment corresponding to the state change amplitude specifically includes: reading the component corresponding to the signal strength change in the average change vector, generating the state change amplitude, where the state change amplitude is a numerical value; when the average change vector contains multiple components corresponding to the signal strength change, averaging the absolute values ​​of the multiple components to generate the state change amplitude; reading the historical minimum and maximum values ​​of the state change amplitude for the same installation point in the point fingerprint database; subtracting the historical minimum value from the current state change amplitude, and then dividing by the difference between the historical maximum and minimum values ​​to generate the normalized state change amplitude; when the historical maximum and minimum values ​​of the state change amplitude are equal, setting the normalized state change amplitude to 0; and multiplying the basic hysteresis threshold by the normalized state change amplitude to generate the threshold increment corresponding to the state change amplitude. The generation of the threshold increment corresponding to the protection impact quantity specifically includes: reading the protection trigger flag and target transmit power level in the radio frequency protection derating instruction; when the protection trigger flag indicates that it has not been triggered, setting the protection impact quantity to 0; when the protection trigger flag indicates that it has been triggered, reading the current transmit power level and target transmit power level, converting the current transmit power level and target transmit power level into level numbers respectively, and subtracting the target transmit power level number from the current transmit power level number to generate the protection impact quantity; when the protection impact quantity is less than 0, setting the protection impact quantity to 0; reading the difference between the maximum allowable transmit power level numbers in the point fingerprint database, dividing the protection impact quantity by the difference between the maximum transmit power level numbers to generate the normalized protection impact quantity; and multiplying the basic hysteresis threshold by the normalized protection impact quantity to generate the threshold increment corresponding to the protection impact quantity. The calibrated signal strength level is generated based on the initial signal strength level, dynamic hysteresis threshold, and radio frequency protective derating command. The signal recognition reliability is then generated based on the calibrated signal strength level and the signal strength state representation. The generation of the calibrated signal strength level specifically includes: reading the initial signal strength level, the signal strength level corresponding to the previous display cycle, and the dynamic hysteresis threshold; calculating the level difference between the level number corresponding to the initial signal strength level and the level number corresponding to the previous display cycle; if the absolute value of the level difference is less than the level change requirement corresponding to the dynamic hysteresis threshold, then the signal strength level corresponding to the previous display cycle is maintained, and the calibrated signal strength level is generated; if the absolute value of the level difference is greater than or equal to the level change requirement corresponding to the dynamic hysteresis threshold, then the initial signal strength level is used as the calibrated signal strength level; when the protection trigger flag in the RF protective derating command indicates that it has been triggered, if the initial signal strength level is higher than the medium signal level, then the calibrated signal strength level is adjusted to the medium signal level; if the initial signal strength level is a weak signal level or a medium signal level, then the current calibrated signal strength level is maintained, thereby keeping the calibrated signal strength level consistent with the RF protection status. The generation of the grade jump threshold specifically includes: reading the dynamic hysteresis threshold, reading the minimum and maximum historical grade jump amplitudes of the same installation point in the point fingerprint database, subtracting the minimum historical grade jump amplitude from the dynamic hysteresis threshold, and then dividing by the difference between the maximum and minimum historical grade jump amplitudes to generate a normalized hysteresis value. When the maximum and minimum historical grade jump amplitudes are equal, the normalized hysteresis value is set to 0. The grade jump threshold is determined based on the normalized hysteresis value. Specifically, when the normalized hysteresis value is less than one-third, the grade jump threshold is set to one grade; when the normalized hysteresis value is greater than or equal to one-third and less than two-thirds, the grade jump threshold is set to two grades; and when the normalized hysteresis value is greater than or equal to two-thirds, the grade jump threshold is set to three grades. The specific steps for generating the calibrated signal strength level are as follows: reading the initial signal strength level, the signal strength level corresponding to the previous display cycle, and the level jump threshold; mapping the weak signal level, medium signal level, good signal level, and excellent signal level to level number one, level number two, level number three, and level number four, respectively; calculating the absolute value of the difference between the level number corresponding to the initial signal strength level and the level number corresponding to the previous display cycle; generating a level difference; when the level difference is less than the level jump threshold, maintaining the signal strength level corresponding to the previous display cycle and generating the calibrated signal strength level; and when the level difference is greater than or equal to the level jump threshold, using the initial signal strength level as the calibrated signal strength level. The generation of signal recognition confidence specifically includes: reading the level number corresponding to the calibrated signal strength level, the signal strength corresponding component in the most recent recursive state, and the signal strength change component in the average change vector; converting the signal strength corresponding component in the most recent recursive state into a state level number; calculating the absolute value of the difference between the state level number and the level number corresponding to the calibrated signal strength level to generate a level deviation; reading the signal strength change component in the average change vector; calculating the degree of deviation of this component relative to the historical average change of the same installation point in the point fingerprint database to generate a state fluctuation deviation; averaging the level deviation and the state fluctuation deviation to generate an identification deviation; and subtracting the identification deviation from 1 to obtain the signal recognition confidence. When this value is less than 0, the signal recognition confidence is set to 0; when this value is greater than 1, the signal recognition confidence is set to 1. The generation of the grade deviation includes: reading the signal strength identification value from the most recent recursive state, reading the weak signal grade boundary, medium signal grade boundary, good signal grade boundary and excellent signal grade boundary corresponding to the current installation point in the point fingerprint database, comparing the signal strength identification value with each grade boundary, generating a state grade number, calculating the absolute value of the difference between the state grade number and the calibrated grade number, and generating the grade deviation. The generation of the state fluctuation deviation includes: reading the component corresponding to the signal strength change in the average change vector as the current signal state change; when there are multiple components corresponding to the signal strength change in the average change vector, calculating the absolute value of each component and averaging the absolute values ​​to generate the current signal state change; reading the historical average change of the same installation point in the point fingerprint database, where the historical average change is obtained by averaging the components of the average change vector corresponding to the signal strength change in the historical upload period of the installation point; calculating the absolute value of the difference between the current signal state change and the historical average change to generate the state change difference; reading the historical maximum change difference of the same installation point in the point fingerprint database; when the historical maximum change difference is not 0, dividing the state change difference by the historical maximum change difference to generate the state fluctuation deviation; when the historical maximum change difference is zero, setting the state fluctuation deviation to 0; when the state fluctuation deviation is greater than 1, setting the state fluctuation deviation to 1; and when the state fluctuation deviation is less than or equal to 1, keeping the state fluctuation deviation unchanged. Indication control parameters are generated based on the calibrated signal strength level, signal identification reliability, and radio frequency protection derating instructions; The generation of the control parameters specifically includes: reading the calibrated signal strength level, signal recognition reliability, and radio frequency protective derating command; generating display color parameters based on the calibrated signal strength level, where weak signal level corresponds to red, medium signal level corresponds to orange, good signal level corresponds to yellow, and excellent signal level corresponds to green; generating brightness constraint parameters based on signal recognition reliability; generating protection prompt parameters based on the radio frequency protective derating command; setting the protection prompt parameters to display when the protection trigger indicator indicates that it has been triggered, and setting the protection prompt parameters to display normally when the protection trigger indicator indicates that it has not been triggered; and generating respiratory cycle parameters based on the calibrated signal strength level and protection prompt parameters. The generation of brightness constraint parameters specifically includes: reading the signal recognition confidence level, which is a value between 0 and 1; reading the normal brightness upper limit and minimum visible brightness corresponding to the current installation point in the point fingerprint database; the normal brightness upper limit is the maximum PWM duty cycle allowed to be output at the installation point under normal display conditions; the minimum visible brightness is the minimum PWM duty cycle required for the on-site identifiable display state; calculating the difference between the normal brightness upper limit and the minimum visible brightness to generate the brightness adjustment range; multiplying the brightness adjustment range by the signal recognition confidence level to generate the confidence brightness increment; and adding the minimum visible brightness to the confidence brightness increment to generate the brightness constraint parameters. The generation of respiratory cycle parameters specifically includes: reading the calibrated signal strength level and protection prompt parameters; assigning weak signal level, medium signal level, good signal level, and excellent signal level to basic respiratory cycles of four seconds, three seconds, two seconds, and one second, respectively; reading the protection prompt parameters; when the protection prompt parameters indicate normal display, using the basic respiratory cycle corresponding to the calibrated signal strength level as the respiratory cycle parameter; when the protection prompt parameters indicate protection prompt display, halving the basic respiratory cycle corresponding to the calibrated signal strength level to generate the respiratory cycle parameter; and when the halved respiratory cycle parameter is less than the minimum prompt cycle, using the minimum prompt cycle as the respiratory cycle parameter. The generation of the minimum prompt period specifically includes: reading the maximum PWM refresh period, the on-site viewing distance, and the on-site ambient brightness level corresponding to the current installation point in the fingerprint database; using 20 times the maximum PWM refresh period as the candidate minimum prompt period; determining the viewing correction period based on the on-site viewing distance: 0.3 seconds when the on-site viewing distance is less than or equal to 1 meter; 0.5 seconds when the on-site viewing distance is greater than 1 meter and less than or equal to 3 meters; and 0.8 seconds when the on-site viewing distance is greater than 3 meters. The brightness correction period is determined based on the on-site ambient brightness level: 0.3 seconds when the on-site ambient brightness level is low; 0.5 seconds when the on-site ambient brightness level is medium; and 0.8 seconds when the on-site ambient brightness level is high. The maximum value among the candidate minimum prompt period, the viewing correction period, and the brightness correction period is determined as the minimum prompt period. The output display status is controlled according to the indicated control parameters, and the calibrated signal strength level, edge transmission status sequence and point status data are associated and written into the point fingerprint database. The signal attenuation fingerprint vector in the point fingerprint database is updated to generate 4GDTU signal strength identification results. The output of the display status specifically includes: controlling the color output of the local indicator unit according to the display color parameters, specifically: red LED is enabled for weak signal level, red and green LEDs are enabled to form an orange display for medium signal level, yellow LED is enabled for good signal level, and green LED is enabled for excellent signal level; controlling the brightness change cycle of the local indicator unit according to the breathing cycle parameters, specifically: dividing the breathing cycle parameters into multiple PWM update times, reading the corresponding brightness coefficient in the sine wave lookup table at each PWM update time, multiplying the brightness coefficient by the brightness constraint parameter to generate the current PWM duty cycle, driving the LEDs corresponding to the display color parameters according to the current PWM duty cycle; controlling the maximum brightness of the local indicator unit according to the brightness constraint parameter, specifically: using the brightness constraint parameter as the upper limit of the PWM duty cycle; when the current PWM duty cycle obtained by multiplying the brightness coefficient output from the sine wave lookup table by the brightness constraint parameter is greater than the brightness constraint parameter, the current PWM duty cycle is set as the brightness constraint parameter; when the current PWM duty cycle is less than or equal to the brightness constraint parameter, the current PWM duty cycle remains unchanged; and controlling the local indicator unit according to the protection prompt parameters. Whether the indicator unit enters the protection prompt display depends on the following: When the protection prompt parameter indicates a normal display, the local indicator unit outputs a continuous breathing display according to the display color parameter, breathing cycle parameter, and brightness constraint parameter. When the protection prompt parameter indicates a protection prompt display, a protection flashing segment is superimposed on the continuous breathing display. The protection flashing segment outputs two short-bright prompts at the beginning of each breathing cycle. The duration of each short-bright prompt is one-tenth of the breathing cycle parameter, and the extinguishing interval between the two short-bright prompts is one-tenth of the breathing cycle parameter. After the short-bright prompt ends, the sinusoidal brightness change within the breathing cycle continues. When the protection prompt parameter indicates a protection prompt display and the target transmit power level in the radio frequency protection derating command is lower than the current transmit power level, the LED brightness constraint parameter corresponding to the display color parameter is reduced to half of the original brightness constraint parameter, and the display color parameter remains unchanged within the protection flashing segment. When the protection prompt parameter indicates a normal display, the protection flashing segment is not executed, and the brightness constraint parameter is not reduced. The display status includes the above-mentioned LED color status, breathing on / off status, maximum brightness status, and protection prompt status. The update of the location fingerprint database specifically includes: reading the location identifier and geographic coordinates from the location status data; establishing a correlation between the calibrated signal strength level and the location identifier and geographic coordinates to generate a location signal level record; reading the upload result, retransmission count, upload delay, and power consumption status from the edge transmission status sequence; associating the location signal level record with the upload result, retransmission count, upload delay, and power consumption status according to the same edge data upload cycle to generate a location update record; and updating the signal attenuation fingerprint vector in the location fingerprint database based on the location update record. The signal attenuation fingerprint vector includes a signal level component, an upload stability component, a retransmission component, an upload delay component, and a power consumption component. The update method is an incremental moving average. Specifically, the old component value of the signal attenuation fingerprint vector and the new component value in the location update record are read, the old component value is multiplied by the moving average retention ratio, the new component value is multiplied by the moving average update ratio, and the two products are added together to obtain the updated component value. The generation of 4G DTU signal strength identification results specifically includes: reading the calibrated signal strength level, signal identification confidence, radio frequency protective derating command, display status, and location fingerprint database update results; reading the protection trigger identifier and target transmit power level from the radio frequency protective derating command; reading the LED color status, breathing on / off status, maximum brightness status, and protection prompt status from the display status; reading the location identifier, geographical coordinates, and updated signal attenuation fingerprint vector from the location fingerprint database update results; generating an identification result record according to the field order of location identifier, geographical coordinates, calibrated signal strength level, signal identification confidence, protection trigger identifier, target transmit power level, LED color status, breathing on / off status, maximum brightness status, protection prompt status, and updated signal attenuation fingerprint vector; and using the identification result record as the 4G DTU signal strength identification result.

[0025] Example 1: To verify the feasibility of this invention in practice, it was applied to a 4G DTU remote data acquisition scenario in a suburban water supply pumping station cluster. The field equipment was distributed in control cabinets along the river, underground valve wells, remote water source wells, and urban pumping stations. The 4G DTUs were used to upload pressure, liquid level, flow rate, and motor operating status. Some installation locations faced issues such as metal cabinet obstruction, underground space attenuation, and coverage edges of base stations. Traditional methods relied solely on CSQ values ​​and network indicator lights, leading to frequent problems such as "network connection displayed but unstable upload" and "signal values ​​briefly normal but increased retransmissions."

[0026] During implementation, the 4G DTU collects edge operation data and reads the location fingerprint database during each edge data upload cycle. After preprocessing, it generates an edge transmission state sequence and location state data. The device extracts service flow features and power consumption features from the sequence to form a service-RF coupling feature vector, which is then input into the edge LTC signal identification model. The model generates an edge signal state vector through an edge construction layer, generates and modulates the signal fluctuation time constant and transmission anomaly time constant through a dual-constant layer, generates a signal strength state representation through a gated recursive layer, and generates indication control parameters by combining RF protective derating instructions at the hysteresis output layer.

[0027] During operation, when the pump station motor starts, causing a sudden increase in data upload content and a greater change in payload, the business flow characteristics can reflect the sudden business surge. When the antenna orientation inside the metal cabinet is poor, leading to increased retransmissions and longer upload delays, the transmission anomaly time constant increases accordingly. When the RF power amplifier current and overall power consumption increase simultaneously, the interlock protection determines and generates an RF protective derating command to prevent the equipment from continuously increasing its transmission power, which would cause a power consumption degradation effect. The final displayed status is jointly determined by the calibrated signal strength level, signal recognition reliability, and protection status, allowing on-site personnel to directly determine whether antenna adjustment or relocation is necessary.

[0028] To illustrate the implementation effect, 4G DTU operation records from the same batch of pumping stations were selected as verification samples, covering installation environments such as control cabinets along rivers, underground valve wells, remote water source wells, and urban pumping stations. The comparison method used the traditional CSQ fixed threshold display method, while the method of this invention uses edge transmission state sequences, service-RF coupling feature vectors, edge LTC signal identification models, interlock protection judgment, and point fingerprint database updates for identification. Evaluation indicators include signal level identification accuracy, upload anomaly identification recall rate, average identification latency, false alarm rate, debugging and positioning time, power consumption anomaly protection trigger accuracy, and point retest consistency rate.

[0029] Table 1 Comparison of 4G DTU signal strength identification and on-site debugging results

[0030] As shown in Table 1, the present invention achieves a signal level identification accuracy of 94.7%, which is 12.3 percentage points higher than the 82.4% accuracy of the traditional CSQ fixed threshold method. This improvement is mainly due to the fact that the edge LTC signal identification model does not rely solely on the instantaneous CSQ value, but instead incorporates the communication state, transmission state, power consumption state, and service-RF coupling changes over a continuous upload cycle into the model. Especially in scenarios such as underground valve wells and metal control cabinets, the CSQ value may remain within the usable range for a short period, but changes in upload latency, retransmission count, and power consumption already reflect a decline in link quality. The present invention can promptly correct the signal level through transmission anomaly time constants and gating recursion.

[0031] The recall rate for upload anomaly identification increased from 76.8% to 92.6%, indicating that this invention has a stronger ability to identify scenarios where "the signal appears usable but the upload is unstable." Traditional methods lack accompanying data collection for upload failures, successful retransmissions, and upload cycle jitter, easily misjudging short-term congestion caused by sudden service interruptions as normal signal levels. This invention uses cross-layer coupled coding based on service flow characteristics and power consumption characteristics, enabling data packet arrival intervals, payload size changes, upload cycle jitter, RF power amplifier current, and overall power consumption to participate in identification. Therefore, it can more accurately distinguish between normal signal fluctuations and accumulated transmission anomalies.

[0032] The average recognition latency was reduced from 18.5s to 7.9s, and the false alarm rate decreased from 13.6% to 5.2%, indicating that the gating recursion layer and the hysteresis output layer simultaneously improved response speed and display stability. The gating recursion layer enhances the absorption capability of the current state when transmission anomalies increase, allowing link degradation to be reflected in the signal strength status representation more quickly; the hysteresis output layer suppresses short-term jumps based on a dynamic hysteresis threshold, preventing frequent color changes in the RGB indicator lights. With both working together, the display status seen by on-site personnel can respond to anomalies promptly without being disturbed by instantaneous fluctuations.

[0033] The average single-point debugging and positioning time was reduced from 21.4 minutes to 12.1 minutes, and the consistency rate of point retests increased from 80.9% to 93.8%, demonstrating the value of updating the point fingerprint database. After on-site personnel move the antenna or adjust the DTU installation position, the system writes the calibrated signal strength level, edge transmission status sequence, and point status data into the point fingerprint database, gradually forming a signal attenuation fingerprint for that installation location. During subsequent retests, the identification process can refer to the historical upload success rate, power consumption baseline, and signal level records of the same point, making the judgment results more consistent with the actual on-site environment.

[0034] The accuracy rate of power consumption anomaly protection triggering increased from 71.5% to 90.3%, indicating that the interlock protection judgment can effectively identify situations where rapid signal fading, link degradation, and abnormal power consumption coexist. Traditional methods typically only indicate weak signals or dropped connections, failing to determine whether continuously increasing transmit power triggers a reverse degradation in power consumption. This invention generates a radio frequency protective derating command when protection conditions are met and incorporates it into the display status control, enabling maintenance personnel to promptly identify points requiring inspection of antenna connections, cabinet heat dissipation, or surrounding obstructions, thereby improving the long-term operational reliability of the 4G DTU.

[0035] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A 4G DTU signal strength identification method based on edge data transmission, characterized in that, include: Collect edge operation data from 4G DTU devices and read the location fingerprint database. After preprocessing, generate edge transmission status sequence and location status data. Service flow features and power consumption features are extracted from the edge transmission state sequence, and cross-layer coupling coding is performed to generate a service-RF coupling feature vector. The edge transmission state sequence and the service-RF coupling feature vector are input into the edge construction layer of the edge LTC signal identification model to generate the edge signal state vector. The edge signal state vector is input into the dual constant layer to generate the signal fluctuation time constant and the transmission anomaly time constant. After mutual modulation, the modulated signal fluctuation time constant and the modulated transmission anomaly time constant are generated. The edge signal state vector, the modulated signal fluctuation time constant, and the modulated transmission anomaly time constant are input into the gated recursive layer to generate transmission gate coefficients. Based on the transmission gate coefficients, the edge signal state vector is gated recursively to generate a signal strength state representation. Based on the modulation signal fluctuation time constant, the modulation transmission abnormality time constant, power consumption characteristics, and point status data, an interlock protection determination is performed, and an RF protective derating instruction is generated. The signal strength status is input to the hysteresis output layer. Based on the signal strength status and the radio frequency protective derating command, the indication control parameters are generated. The display status is output according to the indication control parameters. The location fingerprint database is updated using the calibrated signal strength level, edge transmission status sequence and location status data to generate 4G DTU signal strength identification results.

2. The 4G DTU signal strength identification method based on edge data transmission according to claim 1, characterized in that, The edge operation data includes communication status data, edge transmission result data, and DTU real-time power consumption curve. The point fingerprint database includes point reference data corresponding to the current installation location. The preprocessing includes sampling time correction, invalid record removal, numerical normalization, and time alignment. 3.The 4G DTU signal strength identification method based on edge data transmission according to claim 1, characterized in that, The generation of the service-RF coupling feature vector includes: Extract transmission records from the edge transmission state sequence, and obtain service flow characteristics based on the transmission records; Power consumption change records are extracted from the edge transmission state sequence, and power consumption characteristics are obtained based on the power consumption change records. Wireless link layer features are extracted from the edge transmission state sequence, and service perturbation modulation and power offset correction are applied to the wireless link layer features based on service flow features and power consumption features to generate compensated link features. The service flow characteristics, power consumption characteristics, and compensated link characteristics are combined to generate a service-RF coupling feature vector.

4. The 4G DTU signal strength identification method based on edge data transmission according to claim 1, characterized in that, The generation of the edge signal state vector includes: The edge transmission state sequence and service-RF coupling feature vector are input into the edge construction layer of the edge LTC signal identification model; The edge LTC signal recognition model includes an edge construction layer, a dual constant layer, a gated recursive layer, and a hysteresis output layer; In the edge construction layer, the edge transmission state sequence is temporally encoded to generate an edge temporal state representation; Scale matching is performed on the service-RF coupling feature vector according to the edge temporal state representation to generate a service-RF matching representation; Establish the correspondence between edge timing state representation and service-RF matching representation, and perform state fusion based on the correspondence to generate edge signal state vector.

5. The edge data transmission based 4G DTU signal strength identification method according to claim 1, characterized in that, The generation of the modulated signal fluctuation time constant and the modulated transmission anomaly time constant includes: The edge signal state vector is input into the dual-constant layer, which includes a signal fluctuation constant generation unit, a transmission anomaly constant generation unit, and a dual-constant intermodulation unit. In the fluctuation constant generation unit, the signal change component and link state component in the edge signal state vector are read, and the signal fluctuation time constant is generated according to the signal change amplitude and link state change trend within the continuous upload cycle. The edge signal state vector is input into the transmission anomaly constant generation unit. The transmission change component and power consumption change component in the edge signal state vector are read, and the transmission anomaly time constant is generated according to the duration of the transmission anomaly and the degree of power consumption deviation within the continuous upload cycle. The dual-constant intermodulation unit takes the signal fluctuation time constant and the transmission anomaly time constant as inputs, performs fading enhancement modulation on the transmission anomaly time constant based on the signal fluctuation time constant, and performs anomaly feedback modulation on the signal fluctuation time constant based on the transmission anomaly time constant, to generate the modulated signal fluctuation time constant and the modulated transmission anomaly time constant.

6. The edge data transmission based 4G DTU signal strength identification method according to claim 1, characterized in that, The generation of the signal strength state representation includes: Input the edge signal state vector, the modulated signal fluctuation time constant, and the modulated transmission anomaly time constant into the gated recursive layer; In the gated recursive layer, a recursive input sequence is generated based on the edge signal state vector, the modulation signal fluctuation time constant, and the modulation transmission anomaly time constant. An abnormal absorption coefficient is generated based on the abnormal transmission time constant after modulation, a stability holding coefficient is generated based on the fluctuation time constant of the modulated signal, and a transmission gating coefficient is generated based on the abnormal absorption coefficient and the stability holding coefficient. Candidate signal states are generated based on the recursive input sequence, and the candidate signal states and the recursive states of the previous upload cycle are updated by gating using transmission gating coefficients to generate the recursive states of the current upload cycle. Repeatedly perform gating updates, and fuse the recursive state of the last upload cycle with the recursive state changes in consecutive upload cycles to generate a signal strength state representation.

7. The 4G DTU signal strength identification method based on edge data transmission according to claim 1, characterized in that, The generation of the radio frequency protective derating command includes: Generate signal fading judgment threshold, link degradation judgment threshold, historical power consumption baseline and power consumption protection ratio based on the location status data; The modulation signal fluctuation time constant is compared with the signal fading judgment threshold to generate a signal fading judgment identifier; The abnormal transmission time constant after modulation is compared with the link degradation judgment threshold to generate a link degradation judgment identifier; Generate real-time power deviation indicators based on power consumption characteristics and historical power consumption baseline; Based on the signal fading judgment flag, link degradation judgment flag, and real-time power consumption deviation flag, an RF protective derating command is generated.

8. The 4G DTU signal strength identification method based on edge data transmission according to claim 1, characterized in that, The generation of the 4G DTU signal strength identification result includes: The signal strength state representation is input to the hysteresis output layer, an initial signal strength level is generated based on the signal strength state representation, and a dynamic hysteresis threshold is generated according to the signal strength state representation and the radio frequency protective derating instruction. The calibrated signal strength level is generated based on the initial signal strength level, dynamic hysteresis threshold, and radio frequency protective derating command. The signal recognition reliability is then generated based on the calibrated signal strength level and the signal strength state representation. Indication control parameters are generated based on the calibrated signal strength level, signal identification reliability, and radio frequency protection derating instructions; The system outputs and displays the status according to the indicated control parameters, and associates the calibrated signal strength level, edge transmission status sequence, and location status data into the location fingerprint database. It then updates the signal attenuation fingerprint vector in the location fingerprint database and generates the 4G DTU signal strength identification result.

Images