A big data analysis system based on power bank user usage habit

Abstract

This invention relates to the field of data processing technology and discloses a big data analysis system based on the usage habits of power bank users. It includes a cloud-based big data analysis platform, a cabinet control node, and a power bank communication module. The cloud-based big data analysis platform extracts historical rental characteristics to construct a data increment prediction model, generates predicted data volume labels, and sends them to the cabinet control node. The cabinet control node calculates the available channel evaluation index based on the total predicted data volume and issues power attenuation factor and extended contention window commands when the volume is below a preset congestion threshold. The power bank communication module reduces the gain of the RF front-end amplifier according to the commands and resets the upper limit of the random backoff counter in the carrier sense multiple access (CMI) collision avoidance mechanism to complete data upload. This solution reduces the probability of co-channel interference and the number of data retransmissions, shortens the channel occupancy time, and avoids communication link disconnection due to channel congestion.

Description

Technical Field

[0001] This invention relates to the field of data processing technology and discloses a big data analysis system based on the usage habits of power bank users. Background Technology

[0002] In the field of shared charging equipment operation and maintenance, after a power bank is returned to the cabinet, it needs to upload the stored rental log data to the cabinet's main control node. In existing conventional technical solutions, the communication module inside the power bank and the main control node of the cabinet follow a standard local area network communication protocol. When the cabinet detects the physical access of the power bank and wakes up the communication link, the power bank's communication module directly initiates a connection request at the factory-set default maximum power and uses the fixed minimum contention window value and maximum contention window value in the standard protocol to compete for channel access. The main control node of the cabinet only passively responds to the concurrent upload requests of the power bank as a receiving end and does not interfere with the underlying radio frequency transmission power and protocol contention parameters of the power bank's communication module.

[0003] In scenarios with high-density deployment of power bank cabinets, such as shopping malls and transportation hubs, a large number of power banks are returned in a short period of time. Since all returned power banks simultaneously attempt to access the channel with the default maximum power and fixed contention window parameters, the available resources of the wireless channel are instantly exhausted, resulting in severe co-channel interference and channel conflicts. This dense concurrent access leads to prolonged channel congestion, causing significant data packet loss and repeated retransmissions. Ultimately, the communication link of the cabinet's main control node is disconnected due to congestion timeout, resulting in incomplete transmission of power bank log data. Summary of the Invention

[0004] The purpose of this invention is to provide a big data analysis system based on the usage habits of power bank users, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A big data analysis system based on power bank user habits includes a cloud-based big data analysis platform, a cabinet master control node, and a power bank communication module. The cloud-based big data analysis platform extracts multi-dimensional features from historical power bank rental records, constructs a data increment prediction model for a single rental cycle, generates a predicted data volume label, and sends it to the target cabinet master control node. After detecting a power bank's physical access, the cabinet master control node reads the predicted data volume label and calculates an available channel evaluation index based on the total predicted data volume of currently connected devices. When the available channel evaluation index is lower than a preset congestion threshold, the cabinet master control node sends a power attenuation factor and an extended contention window instruction to the power bank communication module. The power bank communication module reduces the gain of the RF front-end amplifier based on the power attenuation factor and resets the upper limit of the random backoff counter in the carrier sense multiple access (SMI) collision avoidance mechanism based on the extended contention window instruction, completing the log data fragment upload.

[0006] Preferably, the multi-dimensional features extracted from the historical rental records of the power bank by the cloud-based big data analysis platform include usage duration, scanning frequency, and number of abnormal disconnections. The data incremental prediction model inputs the usage duration, scanning frequency, and number of abnormal disconnections into a preset long short-term memory network, and outputs a predicted data volume label corresponding to a single rental period through the fully connected layer of the long short-term memory network. The predicted data volume label includes a data packet quantity field and a single packet byte length field. The cloud-based big data analysis platform sends the predicted data volume label containing the data packet quantity field and the single packet byte length field to the target cabinet master control node.

[0007] Preferably, the rack master control node accumulates the number of data packets in the predicted data volume tag of the currently connected power banks to obtain the total predicted number of data packets. The rack master control node reads the historical statistics of channel idle rate within a preset time period, multiplies the total predicted number of data packets with the historical statistics of channel idle rate to obtain the initial congestion value, divides the initial congestion value by the preset time window length to generate the available channel evaluation index, and retrieves the preset congestion threshold stored locally and compares the available channel evaluation index with the preset congestion threshold.

[0008] Preferably, when the available channel assessment index is lower than the preset congestion threshold, the rack master node calculates the difference between the available channel assessment index and the preset congestion threshold. The rack master node retrieves a pre-stored mapping table in its local storage, and queries the pre-stored mapping table based on the difference to obtain the corresponding power attenuation factor and the extended contention window instruction. The extended contention window instruction includes a minimum contention window parameter and a maximum contention window parameter. The rack master node encapsulates the power attenuation factor, the minimum contention window parameter, and the maximum contention window parameter into a data frame and sends it to the power bank communication module.

[0009] Preferably, the power bank communication module includes a digital-to-analog converter (DAC) and the radio frequency (RF) front-end amplifier. After receiving the power attenuation factor, the power bank communication module performs a logarithmic subtraction operation between the locally stored default transmit power digital signal and the power attenuation factor to generate a target power digital signal. The target power digital signal is then input to the input pin of the DAC. The DAC converts the target power digital signal into a continuous analog voltage signal. The power bank communication module transmits the analog voltage signal to the control pin of the RF front-end amplifier and uses the analog voltage signal to adjust the bias voltage of the RF front-end amplifier to reduce its gain.

[0010] Preferably, the power bank communication module is internally configured with a register, and a default minimum contention window value and a default maximum contention window value are stored in a designated storage area of ​​the register. The power bank communication module parses the extended contention window instruction to obtain the minimum contention window parameter and the maximum contention window parameter, and uses the minimum contention window parameter to overwrite the default minimum contention window value in the designated storage area of ​​the register. At the same time, it uses the maximum contention window parameter to overwrite the default maximum contention window value in the designated storage area of ​​the register. Based on the overwritten default minimum contention window value and the default maximum contention window value, the power bank communication module generates a new random backoff counter upper limit using a random number generator.

[0011] Preferably, the Long Short-Term Memory (LSTM) network includes an input gate, a forget gate, and an output gate. The input gate receives an input vector composed of the usage duration, the scanning frequency, and the number of abnormal disconnections arranged in a time sequence. The forget gate generates a forgetting weight matrix by performing a matrix multiplication operation between the cell state transmitted at the previous time step and the input vector. The output gate outputs a hidden layer state vector by performing a nonlinear transformation between the cell state updated at the current time step and the input vector. The fully connected layer performs a linear transformation and activation function processing on the hidden layer state vector, and outputs the predicted data volume label containing the data packet quantity field and the single packet byte length field.

[0012] Preferably, the storage area of ​​the pre-stored mapping table is divided into multiple sets of difference intervals. Each difference interval corresponds to a set of power attenuation factors, minimum contention window parameters, and maximum contention window parameters. The rack master control node sequentially traverses the multiple sets of difference intervals in the pre-stored mapping table, compares the difference between the available channel evaluation index and the preset congestion threshold to locate the target difference interval, extracts the power attenuation factor, the minimum contention window parameter, and the maximum contention window parameter associated with the target difference interval, and encapsulates the extracted power attenuation factor, the minimum contention window parameter, and the maximum contention window parameter into the extended contention window instruction according to a preset frame format.

[0013] Preferably, the RF front-end amplifier includes a bias current source and a variable resistor network. The analog voltage signal output by the digital-to-analog converter is input to the controlled terminal of the variable resistor network through a wire. The variable resistor network changes its internal equivalent resistance value under the voltage amplitude of the analog voltage signal. The current output by the bias current source flows through the equivalent resistance value to generate the bias voltage. The bias voltage is directly connected to the base or gate of the transistor of the RF front-end amplifier. By changing the voltage value of the bias voltage, the collector current or drain current flowing through the transistor is adjusted to reduce the gain of the RF front-end amplifier.

[0014] Preferably, before overwriting the default minimum contention window value in the specified storage area of ​​the register using the minimum contention window parameter, the power bank communication module retrieves the pre-stored maximum allowed value of the hardware register, compares the minimum contention window parameter with the maximum allowed value of the hardware register, and when the minimum contention window parameter is greater than the maximum allowed value of the hardware register, the power bank communication module uses the maximum allowed value of the hardware register as the actual minimum contention window parameter to overwrite the default minimum contention window value; when the minimum contention window parameter is less than or equal to the maximum allowed value of the hardware register, the power bank communication module directly uses the minimum contention window parameter to overwrite the default minimum contention window value.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention utilizes a cloud-based big data analytics platform to extract usage time, scanning frequency, and abnormal disconnection counts to construct a data increment prediction model. Before the power bank is physically connected, a predicted data volume label containing data packet quantity and single packet byte length fields is sent to the rack control node. The rack control node calculates available channel evaluation indicators based on the total predicted data packet quantity and historical channel idle rate statistics. Before channel congestion occurs, it sends a power attenuation factor and an extended contention window command to the power bank communication module. The power bank communication module, based on these commands, reduces the gain of the RF front-end amplifier and increases the upper limit of the random backoff counter. This reduces the probability of co-channel interference and data retransmissions among multiple devices on the same frequency band from the underlying physical communication mechanism, shortens channel occupancy time, and avoids communication link disconnections caused by prolonged channel congestion.

[0016] 2. During the process of reducing the gain of the RF front-end amplifier, the power bank communication module subtracts the default transmit power digital signal from the power attenuation factor in the logarithmic domain. This subtraction is then converted into an analog voltage signal by a digital-to-analog converter, driving the variable resistor network inside the RF front-end amplifier to change its equivalent resistance value, thereby adjusting the transistor's bias voltage. During the process of resetting the upper limit of the random backoff counter, before writing the minimum contention window parameter to the register, the power bank communication module compares it with the maximum allowed value in the hardware register and overwrites it with the smaller value. This ensures that the actual contention window parameter involved in the calculation is within the physical storage width limit of the hardware register, avoiding abnormal register flipping and communication logic interruption caused by parameter overflow. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall execution process of the present invention. Figure 2 This is a flowchart illustrating the cloud-based prediction model generation process of this invention. Figure 3 This is a flowchart of the cabinet congestion assessment and instruction issuance process of the present invention; Figure 4 This is a flowchart of the gain adjustment process for the RF front-end amplifier of the present invention; Figure 5 This is a flowchart of the random backoff counter upper limit reset process of the present invention; Figure 6 This is a flowchart of the register overwrite boundary security verification process of the present invention. Detailed Implementation

[0018] Please refer to the attached document. Figure 1This embodiment provides a big data analysis system based on power bank user habits, comprising a cloud-based big data analysis platform, a cabinet control node, and a power bank communication module. These three components interact and transmit commands via wireless and wired communication links. The cloud-based big data analysis platform acquires all historical rental records of all power banks through a distributed storage cluster. It extracts multi-dimensional features related to user behavior from these records and constructs an incremental data prediction model using a single rental period as the smallest analytical unit. After the model completes inference calculations, it generates a predicted data volume label for the corresponding power bank. The cloud-based big data analysis platform then distributes the predicted data volume label to the target cabinet control node bound to that power bank via a wide area network communication link.

[0019] The rack master node continuously runs the physical access detection process, identifying the physical access action of the power bank through changes in the state of the hardware level signal. After detecting the physical access of the power bank, the rack master node establishes a data connection with the power bank through a near-field communication link and reads the predicted data volume tag corresponding to the unique identifier of the power bank from the local storage medium. The rack master node aggregates the predicted data volume tags of all connected devices at the current moment to obtain the total predicted data volume of the current channel. Based on the total predicted data volume and the channel's historical operating data, it calculates the available channel evaluation index. At the same time, it retrieves the pre-stored congestion threshold from the local non-volatile storage medium and compares the calculated available channel evaluation index with the preset congestion threshold.

[0020] When the available channel assessment index is lower than the preset congestion threshold, the rack master node generates a corresponding power attenuation factor and an extended contention window command based on the difference between the available channel assessment index and the preset congestion threshold. This power attenuation factor and extended contention window command are then sent to the corresponding power bank's communication module via the established near-field communication link. Upon receiving the command, the power bank's communication module first performs frame parsing and verification. If the verification passes, it adjusts the gain of the RF front-end amplifier based on the power attenuation factor, reducing the RF front-end amplifier's operating gain. Simultaneously, it resets the upper limit of the random backoff counter in the carrier sense multiple access (SMI) collision avoidance mechanism based on the extended contention window command. After completing the parameter configuration, the power bank's communication module performs log data fragmentation and upload operations according to the adjusted RF parameters and channel access parameters.

[0021] The cloud-based big data analytics platform preprocesses the acquired historical rental records using preset data cleaning rules, removing invalid and abnormal data. Invalid data includes rental records with missing fields, and abnormal data includes rental records exceeding the preset reasonable value range. After preprocessing, the platform extracts multi-dimensional features from the valid rental records and arranges these features according to time series to form the input dataset for the incremental data prediction model. The incremental data prediction model achieves parameter convergence through offline training. During the online inference phase, the preprocessed multi-dimensional features are directly input, and the model outputs a predicted data volume label for each rental period. The platform binds each predicted data volume label to a unique device identifier for the corresponding power bank and a unique identifier for the target cabinet. It then distributes the predicted data volume label to the corresponding cabinet master node via address routing. Upon receiving the label, the cabinet master node associates it with the device identifier and stores it in local storage, awaiting the physical access of the corresponding power bank.

[0022] Furthermore, the physical access detection process of the rack master node continuously runs by cyclically scanning the interface level. When a power bank is inserted into the corresponding slot of the rack, the level signal of the slot interface undergoes a preset state change. After the physical access detection process captures this state change, it triggers the subsequent communication link establishment process. The rack master node sends a wake-up command to the power bank through the interface, waking up the communication module inside the power bank and establishing a near-field communication link with the power bank's communication module that conforms to a preset communication protocol. After the link is established, the rack master node sends a device identifier read command to the power bank's communication module to obtain the unique device identifier of the power bank. Based on the device identifier, it retrieves the corresponding predicted data volume tag from the local storage medium. The rack master node maintains a device list of currently connected devices. After each power bank is connected and its tag is read, the predicted data volume tag of that power bank is updated to the device list. Based on the predicted data volume tags of all connected devices in the device list, an aggregation calculation is performed to obtain the total predicted data volume of the current channel.

[0023] After completing the statistics of the total predicted data volume, the rack master node reads the historical statistics of channel idle rate within a preset time period from the local statistics module. Based on the total predicted data volume and the historical statistics of channel idle rate, it performs calculations to generate an available channel evaluation index. The rack master node's storage medium has a pre-allocated fixed storage area for storing a preset congestion threshold. This preset congestion threshold can be remotely updated via a cloud-based big data analysis platform or modified locally via a local configuration interface. The rack master node compares the calculated available channel evaluation index with the preset congestion threshold bit by bit. When the available channel evaluation index is greater than or equal to the preset congestion threshold, the rack master node does not generate control commands and allows the power bank communication module to upload data according to default parameters. When the available channel evaluation index is lower than the preset congestion threshold, the rack master node triggers a parameter control process, calculates the difference between the available channel evaluation index and the preset congestion threshold, generates a corresponding power attenuation factor and an extended contention window command based on the difference, encapsulates the generated parameters and commands into a data frame conforming to a preset communication protocol format, and sends it to the corresponding power bank's communication module.

[0024] The power bank's communication module internally includes an instruction parsing unit, a power control unit, and a channel access control unit. Upon receiving a data frame from the main control node of the cabinet, the instruction parsing unit first performs a cyclic redundancy check on the data frame. If the check passes, the unit parses the data frame, separating the power attenuation factor and the extended contention window instruction. The power attenuation factor is transmitted to the power control unit, and the extended contention window instruction is transmitted to the channel access control unit. Upon receiving the power attenuation factor, the power control unit generates a target power control signal based on a preset power adjustment algorithm and transmits this signal to a digital-to-analog converter (DAC). The DAC converts the digital control signal into an analog voltage signal, which is directly input to the control pin of the RF front-end amplifier (RF front-end amplifier) ​​to adjust its bias voltage, thereby reducing the RF front-end amplifier's gain.

[0025] After receiving the command to extend the contention window, the channel access control unit parses the command to obtain the contention window configuration parameters. Based on the configuration parameters, it overwrites the default contention window value stored in the internal register. Based on the overwritten contention window value, it generates a new random backoff counter upper limit using a random number generator and loads the new random backoff counter upper limit into the corresponding execution unit of the carrier sense multiple access (SMI) collision avoidance mechanism. After completing the configuration of the RF parameters and channel access parameters, the power bank communication module segments the log data to be uploaded according to the preset segmentation rules, adds a corresponding sequence number and verification information to each data segment, and uploads each data segment sequentially according to the configured parameters until all log data segments are uploaded and confirmed.

[0026] In this embodiment, to clarify the standardized definition of multidimensional features in historical rental records and ensure the consistency and processability of input data, the following table is provided: Table 1. Definition of Multidimensional Feature Fields for Historical Power Bank Rental Records The table above defines the field attributes of each multidimensional feature in the historical rental records, clarifies the physical meaning, data format and value rules of each feature, provides standardized input data for the data incremental prediction model, and ensures that the format of the model input data is consistent and the quality is controllable.

[0027] In this embodiment, the cloud-based big data analysis platform is used to extract features of user habits and predict data increments. The predicted data volume tags are distributed before the power bank is physically connected. The main control node of the cabinet is used to quantitatively assess the channel load and identify the congestion status in advance. The power bank's communication module is used to dynamically adjust the radio frequency parameters and channel access parameters, thereby realizing the orderly scheduling of wireless channel access behavior and avoiding channel resource exhaustion and communication link interruption caused by multiple devices accessing concurrently.

[0028] In one alternative embodiment, please refer to the appendix. Figure 2 The cloud-based big data analytics platform extracts multi-dimensional features from the historical rental records of power banks, including usage duration, scanning frequency, and number of abnormal disconnections. The incremental data prediction model is constructed using a long short-term memory (LSTM) network. The cloud-based big data analytics platform arranges the preprocessed usage duration, scanning frequency, and number of abnormal disconnections in a time series to form an input vector. This input vector is then fed into a pre-defined LTM network. The LTM network extracts and fits the input time series features and outputs a predicted data volume label for each rental period via a fully connected layer. The predicted data volume label includes a data packet quantity field and a single packet byte length field. The cloud-based big data analytics platform binds the predicted data volume label, which includes the data packet quantity field and the single packet byte length field, to the corresponding power bank's device identifier and target cabinet identifier, and then sends it to the target cabinet's main control node.

[0029] The cloud-based big data analytics platform sorts each user's historical rental records chronologically, using a single rental period as a single time step. It selects a pre-defined historical time series as the model's input sample. The input vector for each time step consists of the usage duration, scanning frequency, and number of abnormal connection drops corresponding to that rental period. For each input sample, the platform normalizes each feature in the input vector, mapping the value of each feature to a pre-defined numerical range to eliminate dimensional differences between different features. The normalized input vector is then fed into the input layer of a Long Short-Term Memory (LSTM) network. The LSTM network comprises a sequentially connected input layer, hidden layer, and fully connected layer. The hidden layer consists of multiple cascaded LSTM units, each containing an input gate, a forget gate, and an output gate. This gating structure enables long-term memory and short-term updates of time series features, addressing the gradient vanishing and gradient exploding problems inherent in traditional recurrent neural networks.

[0030] Furthermore, the forget gate receives the cell state from the previous time step and the input vector from the current time step. Through linear transformations of the weight matrix and bias vector, and nonlinear transformations of the activation function, it generates a forget weight matrix. Each element in the forget weight matrix ranges from 0 to 1, controlling the degree to which information from the previous cell state is retained. A value of 0 indicates that the information at the corresponding position is completely discarded, while a value of 1 indicates that the information at the corresponding position is completely retained. The input gate receives the hidden layer state vector from the previous time step and the input vector from the current time step. Through linear and nonlinear transformations, it generates an input weight matrix and simultaneously generates candidate cell states for the current time step. The input weight matrix controls the degree to which information from the candidate cell states is written into the current cell state.

[0031] The current cell state is determined by the cell state from the previous time step after processing through the forget gate and the candidate cell state values ​​after processing through the input gate. After updating the cell state, the output gate receives the current input vector and the hidden layer state vector from the previous time step, generates an output weight matrix, and generates the current hidden layer state vector based on the output weight matrix and the updated cell state. The hidden layer state vector output from the last time step of the Long Short-Term Memory network is input to the fully connected layer. The fully connected layer performs a linear transformation and activation function processing on the hidden layer state vector, outputting a 2-dimensional output vector. The two elements of the output vector correspond to the data packet count field and the single packet byte length field, respectively, forming the predicted data volume label.

[0032] In this embodiment, the operations of each layer of the Long Short-Term Memory network are quantitatively described using the following mathematical formula: The formula for the forget gate is: in, Let be the output vector of the forget gate at time t. It is the sigmoid activation function. Here is the weight matrix for the forget gate. for The hidden layer state vector at time step 1. Let be the input vector at time t, which consists of the usage duration, the scanning frequency, and the number of abnormal connection drops. Let be the bias vector of the forget gate. This indicates that two vectors are concatenated.

[0033] The formula for input gate operation is: in, Let be the output vector of the input gate at time t. Here is the weight matrix of the input gate. is the bias vector of the input gate.

[0034] The formula for calculating candidate cell states is: in, Let be the candidate vector of cell states at time t. The hyperbolic tangent activation function is used. This is the weight matrix for cell states. This is the bias vector for the cell state.

[0035] The formula for updating cell state is: in, Let be the cell state vector updated at time t. Let be the cell state vector at time t-1. The Hadamard product operation is the multiplication of corresponding elements of two vectors.

[0036] The formula for the output gate is: in, Let be the output vector of the output gate at time t. This is the weight matrix of the output gate. This is the bias vector for the output gate.

[0037] The formula for the output of the hidden layer state vector is: in, Let be the hidden layer state vector at time t.

[0038] The output formula for a fully connected layer is: in, This is the output vector of the fully connected layer, containing a packet count field and a single packet length in bytes field. It is a linear rectified activation function. This is the weight matrix of the fully connected layer. This is the bias vector for the fully connected layer.

[0039] The Long Short-Term Memory (LSTM) network achieves parameter convergence of the weight matrix and bias vector through offline training. During training, the mean squared error loss function is used as the optimization objective, and the actual number of data packets and the actual byte length of a single packet in the corresponding lease period in the historical lease records are used as label values. The network parameters are iteratively updated through the backpropagation algorithm until the loss function converges to a preset threshold range. After training, the LSM network is deployed to the online inference cluster of a cloud-based big data analytics platform, receiving real-time input user behavior features and outputting corresponding predicted data volume labels.

[0040] Please refer to the attached document. Figure 3 After the cloud-based big data analytics platform generates predicted data volume tags, it adds timestamp information to the tags. The timestamp information corresponds to the valid time range of the predicted rental period. Simultaneously, the data packet quantity field and the single packet byte length field are binary encoded according to preset encoding rules, encapsulated into data packets conforming to the communication protocol format, and sent to the target rack master node via a wide area network link. Upon receiving the predicted data volume tags, the rack master node decodes and verifies the tags. If the verification is successful, the tag is associated with the corresponding power bank device identifier and timestamp information and stored in the local storage medium. At the same time, historical tags exceeding the valid time range are cleaned up to release storage resources.

[0041] In this embodiment, to clarify the structural parameter configuration of the Long Short-Term Memory network and ensure that those skilled in the art can reproduce the data increment prediction model, the following table is provided: Table 2. Parameter configuration of each layer of the Long Short-Term Memory Network Network layer name Input Dimensions Output Dimension Activation function Weight initialization method bias initialization method Input layer 3 3 none none none LSTM hidden layers 3 64 tanh Orthogonal initialization Zero initialization Fully connected layer 64 2 ReLU Gaussian initialization Zero initialization The table above defines the core structural parameters of each layer of the Long Short-Term Memory Network, clarifies the input and output dimensions, activation functions, and parameter initialization rules of each layer, provides clear parameter basis for model reproduction, and ensures that the inference results of the model are reproducible and consistent.

[0042] In this embodiment, the specific implementation of the data increment prediction model is refined. By extracting and fitting the time series features of user habits through a long short-term memory network, the quantitative prediction of the amount of log data generated by the power bank within a single rental period is realized. Through the dual-dimensional output of the data packet quantity field and the single packet byte length field, accurate input basis is provided for subsequent channel load assessment, ensuring that the channel load assessment results match the actual data transmission requirements.

[0043] In one optional embodiment, the rack master node accumulates the number of data packets in the predicted data volume tag of the currently connected power banks to obtain the total predicted number of data packets. The rack master node reads the historical statistics of channel idle rate within a preset time period, multiplies the total predicted number of data packets with the historical statistics of channel idle rate to obtain the initial congestion value, divides the initial congestion value by the preset time window length to generate the available channel evaluation index, and retrieves the preset congestion threshold stored locally and compares the available channel evaluation index with the preset congestion threshold.

[0044] When the available channel assessment index is lower than the preset congestion threshold, the rack master node calculates the difference between the available channel assessment index and the preset congestion threshold. The rack master node retrieves the pre-stored mapping table in local storage and obtains the corresponding power attenuation factor and extended contention window instruction based on the difference. The extended contention window instruction includes the minimum contention window parameter and the maximum contention window parameter. The rack master node encapsulates the power attenuation factor, minimum contention window parameter, and maximum contention window parameter into a data frame and sends it to the power bank communication module.

[0045] Please refer to the attached document. Figure 4 The rack master node maintains a real-time updated list of connected devices. The list stores the device identifier and corresponding predicted data volume tag for each currently connected power bank. Each time a new power bank is connected, the rack master node adds the predicted data volume tag of that power bank to the list of connected devices. Each time a power bank is returned and data is uploaded, the corresponding power bank information is removed from the list of connected devices.

[0046] After each update of the list of connected devices, the rack master node triggers a statistical calculation of the total predicted data packet count. It iterates through all entries in the list of connected devices, extracts the data packet count field of the predicted data volume label in each entry, and sums up the values ​​of all data packet count fields to obtain the total predicted data packet count at the current moment.

[0047] The channel statistics module of the rack master node continuously runs the channel status statistics process, sampling and statistically analyzing the idle and occupied states of the wireless channel according to a preset sampling period. It calculates the channel idle rate for each statistical period, which is the ratio of the duration of channel idleness within that period to the total duration of the statistical period. The channel statistics module stores the calculated channel idle rate in a local historical statistics queue. The historical statistics queue is updated using a first-in, first-out (FIFO) rule, and its length corresponds to the preset statistical time period. After completing the statistics of the total predicted data packets, the rack master node reads all channel idle rate values ​​from the historical statistics queue, calculates the average of all values, and obtains the historical statistical value of the channel idle rate for the preset time period.

[0048] The rack master control node calculates the initial congestion value based on the total predicted number of data packets and the historical statistics of channel idle rate. The formula for calculating the initial congestion value is as follows: in, This is the initial congestion value. This represents the total predicted number of data packets currently connected to the power banks. This refers to the historical statistics of channel idle rate within a preset time period.

[0049] After calculating the initial congestion value, the rack master node calculates the available channel evaluation index based on the initial congestion value and the preset time window length. The formula for calculating the available channel evaluation index is as follows: in, As an indicator for evaluating available channels, The preset time window length corresponds to the time granularity of channel load assessment.

[0050] The rack master node's storage medium has a pre-defined threshold storage area. This area is write-protected, allowing modification only through authorized configuration operations. The area stores a preset congestion threshold, which is the critical value of the available channel assessment index corresponding to a critical congestion state. After calculating the available channel assessment index, the rack master node reads the preset congestion threshold from the threshold storage area, compares the available channel assessment index with the preset congestion threshold, and executes the corresponding operation procedure based on the comparison result.

[0051] When the available channel assessment index falls below the preset congestion threshold, the rack master node triggers a parameter adjustment process. First, it calculates the difference between the preset congestion threshold and the available channel assessment index. This difference characterizes the current channel congestion level; a larger difference indicates a higher risk of channel congestion. The rack master node's storage medium pre-stores a pre-mapped mapping table. The storage area of ​​this table is divided into multiple difference intervals. Each interval corresponds to a power attenuation factor, minimum contention window parameter, and maximum contention window parameter. The difference intervals are arranged in ascending order of the difference; intervals with larger differences correspond to larger power attenuation factor values, as well as larger minimum and maximum contention window parameter values.

[0052] After calculating the difference, the rack master node sequentially traverses multiple difference intervals in the pre-stored mapping table, comparing the calculated difference with the upper and lower limits of each difference interval to locate the target difference interval. Once located, it extracts the power attenuation factor, minimum contention window parameter, and maximum contention window parameter associated with the target difference interval. The rack master node encapsulates the extracted parameters according to a preset frame format. This frame format includes a frame header, device identifier field, power attenuation factor field, minimum parameter field, maximum parameter field, checksum field, and frame trailer. The device identifier field specifies the power bank device corresponding to the parameter, ensuring the parameter only applies to the target device. The checksum field ensures the integrity and correctness of the data frame transmission. After encapsulation, the rack master node sends the data frame to the corresponding power bank's communication module via the established near-field communication link.

[0053] In this embodiment, to clarify the difference interval division and parameter mapping rules of the pre-stored mapping table and to ensure that those skilled in the art can reproduce the parameter scheduling mechanism, the following table is provided: Table 3. Mapping Relationship Between Channel Congestion Difference Range and Communication Parameters The table above defines the communication control parameters corresponding to different congestion difference ranges, clarifies the rules for dividing the difference ranges and the correspondence between the parameters, and provides a clear mapping basis for parameter querying and generation of the rack master control node, ensuring that the control parameters match the actual congestion level of the channel.

[0054] In this embodiment, the calculation logic of available channel assessment indicators and the generation mechanism of congestion control parameters are refined. By combining the total predicted number of data packets with the historical statistical value of channel idle rate, a quantitative assessment of the future load status of the channel is realized. Through the hierarchical difference range and parameter mapping rules, precise hierarchical control of channel access behavior is realized, matching the channel resource scheduling requirements under different congestion risk levels. The parameters are adjusted in advance before channel congestion occurs, thus avoiding the occurrence of channel congestion.

[0055] In one alternative embodiment, please refer to the appendix. Figure 5 The power bank communication module includes a digital-to-analog converter and an RF front-end amplifier. After receiving the power attenuation factor, the power bank communication module performs a logarithmic subtraction operation between the locally stored default transmit power digital signal and the power attenuation factor to generate a target power digital signal. The target power digital signal is then input to the input pin of the digital-to-analog converter, which converts the target power digital signal into a continuous analog voltage signal. The power bank communication module transmits the analog voltage signal to the control pin of the RF front-end amplifier, using the analog voltage signal to adjust the bias voltage of the RF front-end amplifier to reduce its gain.

[0056] The power bank communication module is equipped with registers. The default minimum contention window value and the default maximum contention window value are stored in the designated storage area of ​​the registers. The power bank communication module parses the extended contention window instruction to obtain the minimum contention window parameter and the maximum contention window parameter. It uses the minimum contention window parameter to overwrite the default minimum contention window value in the designated storage area of ​​the register, and at the same time uses the maximum contention window parameter to overwrite the default maximum contention window value in the designated storage area of ​​the register. Based on the overwritten default minimum contention window value and default maximum contention window value, the power bank communication module generates a new random backoff counter upper limit through a random number generator.

[0057] The power control unit of the power bank's communication module stores a default transmit power digital signal in its internal non-volatile storage medium. This default transmit power digital signal corresponds to the maximum transmit power set at the factory default of the power bank's communication module and is stored in the logarithmic field. The unit is quantized. After receiving the power attenuation factor transmitted by the instruction parsing unit, the power control unit performs a subtraction operation in the logarithmic domain on the default transmit power digital signal and the power attenuation factor to generate the target power digital signal. The logarithmic domain operation ensures the linearity and accuracy of power regulation. The formula for calculating the target power digital signal is: in, The logarithmic domain value of the target transmit power, in units of , This is the logarithmic field value of the default transmit power, in units of... , Power attenuation factor, unit: .

[0058] After generating the target power digital signal, the power control unit converts it into binary data that meets the input bit width requirements of the digital-to-analog converter (DAC). This binary data is then input to the DAC's input pins via a parallel data bus. A high-precision reference voltage source is connected to the DAC's reference voltage pin, providing a stable voltage reference for the conversion. Upon receiving the input binary data, the DAC converts the discrete digital signal into a continuous analog voltage signal according to a preset conversion rule. The amplitude of the analog voltage signal is linearly related to the value of the input target power digital signal. The DAC's output pin is connected to the control pin of the RF front-end amplifier via an impedance-matched wire, transmitting the generated analog voltage signal to the RF front-end amplifier's control pin.

[0059] Furthermore, the RF front-end amplifier includes a bias current source, a variable resistor network, and an amplifying transistor. The amplifying transistor is a bipolar transistor or a field-effect transistor. The analog voltage signal output from the digital-to-analog converter is input to the controlled terminal of the variable resistor network through wires. The variable resistor network consists of multiple series and parallel resistor units and switching units. The on / off state of the switching units is controlled by the amplitude of the input analog voltage signal. Driven by the voltage amplitude of the analog voltage signal, the variable resistor network changes the on / off state of its internal switching units, thereby changing the equivalent resistance value inside the variable resistor network. The bias current source is a high-precision constant current source that outputs a constant DC current. The output terminal of the bias current source is connected to one end of the variable resistor network. The constant DC current flows through the equivalent resistance of the variable resistor network, generating a corresponding bias voltage across the variable resistor network. The value of the bias voltage has a linear relationship with the equivalent resistance value of the variable resistor network.

[0060] The output of the variable resistor network is directly connected to the base or gate of the amplifying transistor. The generated bias voltage is directly applied to the base or gate of the amplifying transistor. By changing the bias voltage, the base current or gate voltage of the amplifying transistor is adjusted, thereby adjusting the collector current or drain current flowing through the amplifying transistor and changing its transconductance. The power gain of the RF front-end amplifier is determined by the transconductance of the amplifying transistor and the load resistance, and its calculation formula is as follows: in, The power gain of the RF front-end amplifier, in units of , To amplify the transconductance of the transistor, This is the load resistor for the RF front-end amplifier.

[0061] When the bias voltage decreases, the transconductance of the amplifying transistor decreases accordingly, which in turn leads to a decrease in the power gain of the RF front-end amplifier. This enables precise adjustment of the RF transmission power through the power attenuation factor, reducing the wireless signal coverage of the power bank's communication module and minimizing co-channel interference between multiple devices.

[0062] Please refer to the attached document. Figure 6The channel access control unit of the power bank's communication module is equipped with hardware registers. The designated storage area of ​​these registers is divided into a minimum contention window storage area and a maximum contention window storage area, storing the default minimum and maximum contention window values ​​respectively. These default values ​​are fixed values ​​specified by the communication protocol standard. The register output is directly connected to the execution unit of the carrier sense multiple access (SMI) collision avoidance mechanism, providing parameters for the generation of the random backoff counter. After receiving the extended contention window instruction from the instruction parsing unit, the channel access control unit parses the instruction, extracts the minimum and maximum contention window parameters, and simultaneously retrieves the locally stored maximum allowed value of the hardware register. This maximum allowed value is determined by the register's physical storage width and is the maximum value the register can store.

[0063] The channel access control unit compares the extracted minimum contention window parameter with the maximum allowed value of the hardware register. When the minimum contention window parameter is greater than the maximum allowed value, the channel access control unit uses the maximum allowed value as the actual minimum contention window parameter and overwrites the default minimum contention window value in the register's minimum contention window storage area. When the minimum contention window parameter is less than or equal to the maximum allowed value, the channel access control unit directly uses the extracted minimum contention window parameter to overwrite the default minimum contention window value in the register's minimum contention window storage area. The limit handling formula for the minimum contention window parameter is: in, This is the minimum contention window parameter actually used for overwriting. The minimum contention window parameter received. This is the maximum allowed value for the hardware register.

[0064] The channel access control unit uses the same processing logic to perform limit processing and overwrite operations on the maximum contention window parameter, ensuring that the parameter value written to the register is within the physical storage bit width limit of the hardware register. After completing the register overwrite operation, the channel access control unit triggers the update process of the random backoff counter upper limit. Based on the overwritten minimum and maximum contention window values, a hardware random number generator generates a uniformly distributed random number. The value of the random number ranges from the overwritten minimum to maximum contention window values. The generated random number serves as the new random backoff counter upper limit and is loaded into the execution unit of the carrier sense multiple access (SMI) collision avoidance mechanism. The formula for generating the random backoff counter upper limit is: in, This is the upper limit of the random backoff counter. For uniformly distributed hardware random number generation functions, This is the minimum contention window size after overwriting. This represents the maximum contention window value after overwriting.

[0065] The execution unit of the Carrier Sense Multiple Access (CSMA) collision avoidance mechanism first listens to the channel state when performing channel access operations. When it detects that the channel is busy, it waits for the channel to become idle. After the channel becomes idle, the upper limit of the generated random backoff counter is loaded into the backoff counter. The backoff counter decrements according to a preset time step while the channel remains idle. When the backoff counter decrements to 0, the data transmission operation is performed. If the channel becomes busy during the backoff counter decrement process, the decrement operation is paused, and the decrement continues after the channel becomes idle again. By increasing the value range of the random backoff counter, the probability of multiple devices transmitting data simultaneously is reduced, thereby reducing channel collisions and data retransmissions.

[0066] In this embodiment, to clarify the gain adjustment characteristics of the RF front-end amplifier and ensure that those skilled in the art can reproduce the power adjustment mechanism, the following table is provided: Table 4. Relationship between RF front-end amplifier gain and bias voltage Bias voltage (mV) Equivalent resistance (kΩ) of a variable resistor network Collector current (mA) Transistor transconductance (mS) Power gain (dB) 800 1 10 38.5 22 700 1.2 8 30.8 20 600 1.5 6 23.1 17 500 2 4 15.4 14 400 3 2 7.7 8 The table above defines the correspondence between the bias voltage and equivalent resistance, collector current, transistor transconductance, and power gain of the RF front-end amplifier, and clarifies the linear relationship between the analog voltage signal and the gain adjustment amount, providing a clear basis for the precise adjustment of RF power.

[0067] In this embodiment, the underlying hardware and logic implementation of RF power adjustment and contention window reset are refined. Through power calculation in the digital domain and digital-to-analog conversion, continuous and precise control of transmit power is achieved. Through the adjustment of variable resistor network and transistor bias voltage, direct mapping from digital instructions to hardware gain adjustment is realized. Through limit verification of hardware register parameters, abnormal register flipping and communication logic interruption caused by parameter overflow are avoided. Orderly management of wireless channel access behavior is realized from two dimensions: physical layer and medium access control layer, reducing the probability of channel conflict and co-channel interference.

Claims

1. A big data analysis system based on power bank user habits, characterized in that, The system includes a cloud-based big data analytics platform, a cabinet master control node, and a power bank communication module. The cloud-based big data analytics platform extracts multi-dimensional features from the historical rental records of the power banks, constructs a data increment prediction model for a single rental period, generates a predicted data volume label, and sends it to the target cabinet master control node. After detecting the physical access of the power bank, the cabinet master control node reads the predicted data volume label and calculates the available channel evaluation index based on the total predicted data volume of the currently connected devices. When the available channel evaluation index is lower than a preset congestion threshold, the cabinet master control node sends a power attenuation factor and an extended contention window instruction to the power bank communication module. The power bank communication module reduces the gain of the RF front-end amplifier according to the power attenuation factor and resets the upper limit of the random backoff counter in the carrier sense multiple access collision avoidance mechanism according to the extended contention window instruction, completing the log data fragment upload.

2. The big data analysis system based on power bank user habits according to claim 1, characterized in that, The multidimensional features extracted from the historical rental records of the power banks by the cloud-based big data analysis platform include usage duration, scanning frequency, and number of abnormal disconnections. The incremental data prediction model inputs the usage duration, scanning frequency, and number of abnormal disconnections into a preset long short-term memory network, and outputs a predicted data volume label corresponding to a single rental period through the fully connected layer of the long short-term memory network. The predicted data volume label includes a data packet quantity field and a single packet byte length field. The cloud-based big data analysis platform sends the predicted data volume label containing the data packet quantity field and the single packet byte length field to the target rack master node.

3. The big data analysis system based on power bank user habits according to claim 2, characterized in that, The rack master control node accumulates the number of data packets in the predicted data volume tag of the currently connected power banks to obtain the total predicted number of data packets. The rack master control node reads the historical statistics of channel idle rate within a preset time period, multiplies the total predicted number of data packets with the historical statistics of channel idle rate to obtain the initial congestion value, divides the initial congestion value by the preset time window length to generate the available channel evaluation index, and retrieves the preset congestion threshold stored locally and compares the available channel evaluation index with the preset congestion threshold.

4. The big data analysis system based on power bank user habits according to claim 3, characterized in that, When the available channel assessment index is lower than the preset congestion threshold, the rack master node calculates the difference between the available channel assessment index and the preset congestion threshold. The rack master node retrieves the pre-stored mapping table in its local storage and queries the pre-stored mapping table based on the difference to obtain the corresponding power attenuation factor and the extended contention window instruction. The extended contention window instruction includes minimum contention window parameters and maximum contention window parameters. The rack master node encapsulates the power attenuation factor, the minimum contention window parameters, and the maximum contention window parameters into a data frame and sends it to the power bank communication module.

5. The big data analysis system based on power bank user habits according to claim 1, characterized in that, The power bank communication module includes a digital-to-analog converter (DAC) and an RF front-end amplifier. After receiving the power attenuation factor, the power bank communication module performs a logarithmic subtraction operation between the locally stored default transmit power digital signal and the power attenuation factor to generate a target power digital signal. The target power digital signal is then input to the input pin of the DAC. The DAC converts the target power digital signal into a continuous analog voltage signal. The power bank communication module transmits the analog voltage signal to the control pin of the RF front-end amplifier to adjust the bias voltage of the RF front-end amplifier.

6. The big data analysis system based on power bank user habits according to claim 1, characterized in that, The power bank communication module is internally configured with a register. A default minimum contention window value and a default maximum contention window value are stored in a designated storage area of ​​the register. The power bank communication module parses the extended contention window instruction to obtain the minimum contention window parameter and the maximum contention window parameter. The minimum contention window parameter is used to overwrite the default minimum contention window value in the designated storage area of ​​the register. At the same time, the maximum contention window parameter is used to overwrite the default maximum contention window value in the designated storage area of ​​the register. Based on the overwritten default minimum contention window value and the default maximum contention window value, the power bank communication module generates a new random backoff counter upper limit using a random number generator.

7. The big data analysis system based on power bank user habits according to claim 2, characterized in that, The Long Short-Term Memory (LSTM) network includes an input gate, a forget gate, and an output gate. The input gate receives an input vector composed of the usage duration, the scanning frequency, and the number of abnormal disconnections arranged in a time sequence. The forget gate generates a forgetting weight matrix by performing a matrix multiplication operation between the cell state transmitted at the previous time step and the input vector. The output gate performs a nonlinear transformation between the cell state updated at the current time step and the input vector to output a hidden layer state vector. The fully connected layer performs a linear transformation and activation function processing on the hidden layer state vector to output the predicted data volume label containing the data packet quantity field and the single packet byte length field.

8. The big data analysis system based on power bank user habits according to claim 4, characterized in that, The storage area of ​​the pre-stored mapping table is divided into multiple sets of difference intervals. Each difference interval corresponds to a set of power attenuation factors, minimum contention window parameters, and maximum contention window parameters. The rack master control node sequentially traverses the multiple sets of difference intervals in the pre-stored mapping table, compares the difference between the available channel evaluation index and the preset congestion threshold to locate the target difference interval, extracts the power attenuation factor, minimum contention window parameter, and maximum contention window parameter associated with the target difference interval, and encapsulates the extracted power attenuation factor, minimum contention window parameter, and maximum contention window parameter into the extended contention window instruction according to a preset frame format.

9. The big data analysis system based on power bank user habits according to claim 5, characterized in that, The RF front-end amplifier includes a bias current source and a variable resistor network. The analog voltage signal output by the digital-to-analog converter is input to the controlled terminal of the variable resistor network through a wire. The variable resistor network changes its internal equivalent resistance value under the voltage amplitude of the analog voltage signal. The current output by the bias current source flows through the equivalent resistance value to generate the bias voltage. The bias voltage is directly connected to the base or gate of the transistor of the RF front-end amplifier. The collector current or drain current flowing through the transistor is adjusted by changing the voltage value of the bias voltage.

10. The big data analysis system based on power bank user habits according to claim 6, characterized in that, Before overwriting the default minimum contention window value in the specified storage area of ​​the register using the minimum contention window parameter, the power bank communication module retrieves the pre-stored maximum allowed value of the hardware register and compares the minimum contention window parameter with the maximum allowed value of the hardware register. When the minimum contention window parameter is greater than the maximum allowed value of the hardware register, the power bank communication module uses the maximum allowed value of the hardware register as the actual minimum contention window parameter to overwrite the default minimum contention window value. When the minimum contention window parameter is less than or equal to the maximum allowed value of the hardware register, the power bank communication module directly uses the minimum contention window parameter to overwrite the default minimum contention window value.

Images