A method for monitoring the status of a concentrator
By using a BP neural network prediction model, combined with the real-time communication status of the concentrator and the number of meters, the problem of real-time monitoring of the concentrator status was solved, achieving accurate real-time monitoring of the concentrator status, reducing the risk of meter reading failure, and improving the accuracy and efficiency of the remote meter reading system.
Patent Information
- Application Number
- CN202411179475.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-27
AI Technical Summary
Existing technologies cannot achieve real-time monitoring of the concentrator's status, which increases the risk of meter reading failure in remote meter reading systems.
A response time prediction model based on a BP neural network is used to predict the standard response time of the heartbeat signal by combining the real-time communication status of the concentrator and the number of meters that need to be read at the next moment. The status of the concentrator is then determined by the response of the heartbeat signal.
It enables real-time monitoring of the concentrator's communication status, reduces the risk of meter reading failure, and improves the accuracy and efficiency of the remote meter reading system.
Smart Images

Figure CN118714048B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of power concentrators, and in particular to a state monitoring method for a concentrator. Background Art
[0002] Remote centralized meter reading system is a development trend in the fields of water, electricity, coal and other industrial control in recent years. In the power field, the number of electricity meters is increasing day by day, and remote centralized meter reading technology can remotely and automatically read the readings of multiple meters. In the power field, the remote centralized meter reading system includes multiple meter terminals, collectors, concentrators and intelligent master stations, and a concentrator is connected to a multi-level collector, and the multi-level collector is connected to the meter terminal. The concentrator is mainly responsible for sending meter reading instructions to the meter terminal through the multi-level collector, thereby reading the data of multiple meter terminals and realizing remote centralized meter reading; in order to ensure the accuracy of remote centralized meter reading, it is necessary to monitor the status of the concentrator in real time to avoid meter reading failure due to abnormal communication status.
[0003] At present, a patent application document with publication number CN114339478A discloses a communication fault interval detection method and system for a concentrator, the method comprising: performing a basic performance test on a first concentrator to obtain a first test result; based on the first test result, obtaining a first concentrator set, performing real-time regional division on the first concentrator set through a first regional segmentation model to obtain a first real-time regional ID; obtaining a first uplink communication link signal acquisition result based on the first real-time regional ID; inputting the first collector set into a second regional segmentation model to obtain a first collector set segmentation regional ID set; obtaining a first downlink communication link signal acquisition result based on the first collector set segmentation regional ID set; and performing abnormal interval detection on the communication fault of the first collector based on the first uplink communication link signal acquisition result and the first downlink communication link signal acquisition result.
[0004] The above method can detect abnormal intervals of communication failures in the first collector. However, basic performance testing of the first concentrator is still required. Basic performance testing needs to be completed by manual regular inspections, and real-time monitoring of the communication status of the concentrator cannot be achieved. Summary of the invention
[0005] In order to solve the technical problem that it is impossible to realize real-time monitoring of the status of the concentrator, the present application provides a status monitoring method for the concentrator, which can obtain accurate status monitoring results and realize real-time monitoring of the communication status of the concentrator.
[0006] The present application provides a state monitoring method for a concentrator, which is applied to an intelligent main station or any collector of the concentrator. The monitoring method comprises: inputting the real-time communication state of the concentrator and the number of meters of each type of electricity consumption that needs to be read at the next moment into a trained response time prediction model, predicting the standard response time of a heartbeat signal at the next moment, the electricity consumption types include residential electricity consumption, industrial electricity consumption, commercial electricity consumption and public facility electricity consumption, and the real-time communication state includes the instantaneous bandwidth occupancy rate of each channel; sending a heartbeat signal at the next moment, and if no confirmation signal is received after responding to the interval standard response time, the state monitoring result of the concentrator is abnormal, otherwise, the state monitoring result is normal; the response time prediction model is a BP neural network.
[0007] Collect the real-time communication status of the concentrator at the current moment, and the number of meters of each type of electricity consumption that need to be read at the next moment; input the number of meters of each type of electricity consumption that need to be read at the next moment, and the real-time communication status into the trained response time prediction model to accurately predict the standard response time of the heartbeat signal at the next moment; send the heartbeat signal at the next moment. If the confirmation signal is not received within the standard response time, it means that the status monitoring result of the concentrator is abnormal, thereby realizing real-time monitoring of the concentrator communication status.
[0008] Preferably, the response time prediction model includes an input layer, multiple hidden layers and an output layer; the input layer is used to receive the real-time communication status and the number of electricity meters of each electricity consumption type; the multiple hidden layers are used to perform dimensionality transformation on the concatenated vector composed of the real-time communication status and the number of electricity meters of each electricity consumption type; the output layer is used to regress the output result of the last hidden layer to the standard response time; the input layer and each hidden layer include multiple neurons, and the activation function of each neuron is a nonlinear activation function; the output layer includes 1 neuron, and the activation function of the neuron is a ReLU function.
[0009] An end-to-end model structure of a response time prediction model is proposed. The activation function of neurons in the input layer and each hidden layer adopts a nonlinear activation function, which effectively improves the learning ability of the response time prediction model and ensures that the response time prediction model can learn linear and nonlinear features. The activation function of neurons in the output layer adopts the ReLU function to ensure that the output standard response time is a positive number.
[0010] Preferably, the nonlinear activation function includes a sigmoid function and a tanh function.
[0011] Preferably, the instantaneous bandwidth occupancy of the channel includes the bandwidth occupancy of the uplink and the bandwidth occupancy of the downlink.
[0012] During the operation of the concentrator, the concentrator needs to first send a meter reading instruction to the meter. After receiving the meter reading instruction, the meter transmits the meter reading information to the concentrator. Therefore, each channel is a two-way communication. The bandwidth occupancy rate of the uplink and downlink of each channel is collected to accurately reflect the communication status of each channel.
[0013] Preferably, the heartbeat signal includes preset information; if the time interval in response to receiving a confirmation signal is less than or equal to the standard response duration, and the preset information in the confirmation signal is the same as the preset information of the heartbeat signal, the status monitoring result of the concentrator is normal; otherwise, the status monitoring result is abnormal.
[0014] Preferably, the training method of the response time prediction model: collects the communication status at a historical moment and the number of electricity meters of each type of electricity consumption that needs to be read at the next moment after the historical moment as training samples, and collects the actual response time of each training sample; clusters the training samples to obtain multiple clusters, performs anomaly detection on the actual response time of the training samples in each cluster, and obtains the time anomaly of each training sample; calculates the selection probability of each training sample, the selection probability is negatively correlated with the time anomaly, and is positively correlated with the number of training samples in the cluster to which the training sample belongs; selects multiple batches of training samples based on the selection probability, and trains the response time prediction model using mean square error loss to obtain a trained response time prediction model.
[0015] Taking into account that the actual response time corresponding to the training samples may be collected under abnormal conditions of the concentrator, the selection probability of each training sample is calculated according to the clustering algorithm and the anomaly detection algorithm. When selecting multiple batches of training samples, training samples with a large selection probability are given priority to ensure that the trained response time prediction model can output accurate standard response time.
[0016] Preferably, the training samples The probability of selection Satisfies the relationship:
[0017] , For training samples The number of training samples in the cluster to which it belongs, is the total number of training samples, For training samples The duration of abnormality.
[0018] The duration anomaly can reflect the degree of abnormality of the actual response duration corresponding to the training sample. The larger the duration anomaly, the more the actual response duration of the training sample deviates from the normal state. The smaller the number of training samples in a cluster, the less accurate the duration anomaly of each training sample in the cluster, and the greater the possibility that the concentrator state corresponding to the training sample is an abnormal state. Therefore, the smaller the duration anomaly, the larger the number of training samples in the cluster, and the greater the selection probability.
[0019] Preferably, clustering the training samples comprises: clustering the training samples based on a density clustering algorithm, wherein the density clustering algorithm is a DBSCAN algorithm or a HDBSCAN algorithm.
[0020] The technical solution of this application has the following beneficial technical effects:
[0021] The above-mentioned state monitoring method of the concentrator provided in the embodiment of the present application collects the real-time communication state of the concentrator at the current moment, and the number of electricity meters of each type of electricity consumption that needs to be read at the next moment, wherein the number of electricity meters of each type of electricity consumption that needs to be read directly determines the amount of data that needs to be transmitted at the next moment; the number of electricity meters of each type of electricity consumption that needs to be read at the next moment, and the real-time communication state are input into the trained response time prediction model to accurately predict the standard response time of the heartbeat signal at the next moment; a heartbeat signal is sent at the next moment, and if a confirmation signal is not received within the standard response time, it means that the state monitoring result of the concentrator is abnormal, thereby realizing real-time monitoring of the communication state of the concentrator. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easy to understand. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:
[0023] Figure 1 is a flow chart of a method for monitoring the status of a concentrator according to an embodiment of the present application;
[0024] Figure 2 is a schematic diagram of each channel of a concentrator according to an embodiment of the present application;
[0025] Figure 3 It is a flowchart of a response time prediction model training method according to an embodiment of the present application. DETAILED DESCRIPTION
[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.
[0027] It should be understood that when the application uses the terms "first", "second", etc., they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the application indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.
[0028] The present application provides a state monitoring method for a concentrator, which is applied to the intelligent master station or any collector of the concentrator, and can monitor the communication state of the concentrator in real time. In a remote meter reading system, a large number of electric meters can be connected to a concentrator through multiple collectors, and each electric meter corresponds to its own meter reading cycle; the concentrator plays a connecting role in the remote meter reading system. After sending meter reading instructions to the electric meter according to the meter reading cycle of different electric meters, the electric meter transmits the meter reading information back to the concentrator through the collector; after the concentrator receives the meter reading information, it transmits the meter reading information to the intelligent master station for storage and analysis. Therefore, the communication state of the concentrator directly affects the meter reading efficiency and accuracy of the remote meter reading system.
[0029] At present, in order to judge the communication status of the concentrator, the intelligent master station sends a heartbeat signal to the concentrator at a preset interval. After receiving the heartbeat signal, the concentrator will send a confirmation signal back to the intelligent master station. The time interval between the sending of the heartbeat signal and the receiving of the confirmation signal by the intelligent master station is the response time of the heartbeat signal. If the intelligent master station still does not receive the confirmation signal after a fixed time interval, that is, the response time is greater than the fixed time, it means that the status monitoring result of the concentrator is abnormal. However, due to the different number of meters that need to be read at different times, the communication capabilities of the concentrators at different times are inconsistent. Using a fixed time to judge the status monitoring result will inevitably produce errors, resulting in inaccurate status monitoring results.
[0030] See also Figure 1 FIG. 1 is a flow chart of a method for monitoring the status of a concentrator according to an embodiment of the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted.
[0031] S11, input the real-time communication status of the concentrator and the number of meters of each type of electricity consumption that need to be read at the next moment into the trained response time prediction model to predict the standard response time of the heartbeat signal at the next moment, wherein the electricity consumption types include residential electricity, industrial electricity, commercial electricity and public facility electricity, and the real-time communication status includes the instantaneous bandwidth occupancy of each channel.
[0032] In one embodiment, among the multiple electricity meters connected to the concentrator, the meter reading cycles of different electricity meters are known information, and the electricity meters that need to be read at any time can be determined based on the meter reading cycle of each electricity meter; further, since the electricity usage types in the areas where different electricity meters are located are different, the electricity meters can be divided into multiple types of electricity meters: residential electricity meters, industrial electricity meters, commercial electricity meters and public facility electricity meters, and the data volume of meter reading information of different types of electricity meters during meter reading operations is different.
[0033] Specifically, for residential electricity meters, the meter reading operation is mainly used to monitor the residents' electricity usage, and the meter reading information mainly includes the total electricity consumption; while for public facility electricity meters (such as street lighting, traffic signals and hospitals), while monitoring the electricity usage, it focuses more on the real-time acquisition of dynamic data of public facilities, fault warning, etc., and the meter reading information during the meter reading operation includes not only the total electricity consumption, but also other electricity parameters such as power factor, power and frequency; therefore, the number of meters of each type of electricity consumption that need to be read at the next moment can affect the amount of data that needs to be transmitted at the next moment, affect the communication load of the concentrator, and then affect the standard response time at the next moment, where the standard response time is the response time of the concentrator to the heartbeat signal under normal conditions.
[0034] The real-time communication status of the concentrator includes the instantaneous bandwidth occupancy rate of each channel, which can reflect the usage of the concentrator's communication resources at the current moment; the trained response time prediction model comprehensively considers the usage of the concentrator's communication resources at the current moment and the number of each type of meter that needs to be read at the next moment to predict the standard response time of the heartbeat signal at the next moment.
[0035] In one embodiment, the response time prediction model is a BP neural network. Specifically, the response time prediction model includes an input layer, multiple hidden layers and an output layer; the input layer is used to receive the real-time communication status and the number of electric meters of each type of electricity consumption; the multiple hidden layers are used to perform dimension transformation on the concatenated vector composed of the real-time communication status and the number of electric meters of each type of electricity consumption; the output layer is used to regress the output result of the last hidden layer into a standard response time.
[0036] Among them, the input layer and each hidden layer include multiple neurons, and the activation function of each neuron is a nonlinear activation function, which includes a sigmoid function and a tanh function, which are used to introduce nonlinear mapping and improve the learning ability of the BP neural network, so that the response time prediction model can learn the nonlinear relationship between the standard response time and the input information; the output layer includes 1 neuron, and the activation function of the neuron is a ReLU function, which is used to ensure that the output standard response time is greater than 0.
[0037] It should be noted that the dimension of the real-time communication state is recorded as A, and the number of electricity types is recorded as B. Then the number of neurons in the input layer is , to ensure that the input layer can receive all input information; the number of hidden layers and the number of neurons contained in each hidden layer are not limited in this application.
[0038] For example, see Figure 2 , is a schematic diagram of each channel of the concentrator according to an embodiment of the present application; there are channels for two-way communication between the concentrator and one intelligent master station and five collectors. By collecting the instantaneous bandwidth occupancy of each channel at the current moment, the real-time communication status can be obtained. The real-time communication status is a vector of 6 rows and 1 column, that is, the dimension of the real-time communication status is 6; and the electricity consumption types include 4 categories: residential electricity, industrial electricity, commercial electricity and public facility electricity, then the number of neurons in the input layer is 10.
[0039] In another embodiment, since there are bidirectional communication channels between the concentrator and one intelligent master station and five collectors, that is, each channel includes an uplink and a downlink, wherein the uplink is a link from the concentrator to the intelligent master station or collector, and the downlink is a link from the intelligent master station or collector to the concentrator, and the communications of the uplink and the downlink are independent of each other, the instantaneous bandwidth occupancy of the channel includes the bandwidth occupancy of the uplink and the bandwidth occupancy of the downlink. For example, please continue to refer to Figure 2 For each channel of the concentrator shown, one channel corresponds to 2 bandwidth occupancy rates, and the real-time communication state is a vector of 12 rows and 1 column, that is, the dimension of the real-time communication state is 12; and the electricity consumption types include 4 categories: residential electricity, industrial electricity, commercial electricity and public facility electricity, then the number of neurons in the input layer is 16.
[0040] It can be understood that since the response time of the heartbeat signal includes the time interval from the smart master station sending the heartbeat signal to the concentrator, and the time interval from the concentrator sending the confirmation signal to the smart master station, that is, the response time of the heartbeat signal involves both the uplink and the downlink. Therefore, the bandwidth occupancy rate of the uplink and downlink of each channel is taken into account in the real-time communication status, the amount of information received and sent by the concentrator at the current moment can be accurately obtained, and the standard response time of the heartbeat signal at the next moment can be accurately predicted.
[0041] In one embodiment, the real-time communication status of the concentrator at the current moment and the number of meters of each type of electricity consumption that need to be read at the next moment are input into the trained response time prediction model to output the standard response time of the heartbeat signal at the next moment.
[0042] In this way, with the help of the trained response time prediction model, the standard response time of the heartbeat signal at the next moment can be predicted. The standard response time takes into account the communication status of the concentrator at the current moment and the impact of the meter reading task to be performed on the information transmission speed. It is the response time of the heartbeat signal of the concentrator under normal conditions.
[0043] S12, sending a heartbeat signal at the next moment, if no confirmation signal is received after the standard response time interval, the status monitoring result of the concentrator is abnormal, otherwise, the status monitoring result is normal.
[0044] In one embodiment, after obtaining the standard response time of the heartbeat signal at the next moment, the intelligent master station or any collector sends a heartbeat signal to the concentrator at the next moment. After the heartbeat signal is transmitted to the concentrator, the concentrator transmits the received heartbeat signal back to the intelligent master station or any collector as a confirmation signal; if the confirmation signal is not received after the standard response time, it indicates an abnormal situation such as communication delay or communication interruption. The status monitoring result of the concentrator at the next moment is marked as abnormal, and an alarm is issued.
[0045] On the contrary, if the time interval for receiving the confirmation signal is less than or equal to the standard response time, the status monitoring result of the concentrator is normal.
[0046] It should be noted that the intelligent master station and the collector and concentrator can all achieve two-way communication. Therefore, the heartbeat signal can be sent by the intelligent master station or by any collector, and this application does not impose any restrictions.
[0047] In another embodiment, the heartbeat signal includes preset information; in response to the time interval of receiving the confirmation signal being less than or equal to the standard response duration, and the preset information in the confirmation signal is the same as the preset information of the heartbeat signal, the status monitoring result of the concentrator is normal, otherwise, the status monitoring result of the concentrator is abnormal; the preset information can be a specific numerical value to simulate the meter reading information of the electric meter during the meter reading operation.
[0048] It can be understood that the time interval for receiving the confirmation signal is less than or equal to the standard response time, indicating that there is no communication delay or communication interruption; the preset information in the confirmation signal is the same as the preset information of the heartbeat signal, indicating that there is no information missing; at this time, the status monitoring result of the concentrator is marked as normal.
[0049] In one embodiment, in the above concentrator condition monitoring method, the prediction accuracy of the response time prediction model directly affects the accuracy of the condition monitoring result. Figure 3 , is a flow chart of a response duration prediction model training method according to an embodiment of the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted, which is described in detail as follows.
[0050] S21, collecting the communication status at any historical moment and the number of electricity meters of each type that need to be read at the next moment of the historical moment as training samples, and collecting the actual response time of each training sample.
[0051] Among them, a heartbeat signal is sent to the concentrator at every moment, multiple groups of training samples can be collected in the historical time period, and the actual response time of the heartbeat signal corresponding to each group of training samples can be collected.
[0052] The actual response duration of each training sample is the actual response duration of the heartbeat signal at the next moment of the historical moment.
[0053] S22, clustering the training samples to obtain a plurality of clusters, performing anomaly detection on the actual response duration of the training samples in each cluster, and obtaining the duration anomaly degree of each training sample.
[0054] The DBSCAN algorithm or HDBSCAN algorithm based on density clustering can be used to cluster the training samples to obtain multiple clusters, and each cluster includes at least one training sample. The DBSCAN algorithm or HDBSCAN algorithm does not need to pre-set the number of clusters, and can automatically determine the number of clusters according to the data distribution of the training samples.
[0055] In a cluster, the similarity between its training samples is relatively high, so the actual response time of the training samples in a cluster should remain consistent or have little difference; therefore, the actual response time of the training samples in each cluster is detected for anomalies using the isolation forest or LOF anomaly detection algorithm to obtain the duration anomaly of each training sample.
[0056] In this way, the duration abnormality can reflect the degree of abnormality of the actual response duration corresponding to the training sample. The larger the duration abnormality, the more the actual response duration of the training sample deviates from the normal state, and the smaller the role of the training sample in the training process of the response duration prediction model.
[0057] S23, calculating the selection probability of each training sample, wherein the selection probability is negatively correlated with the duration abnormality and positively correlated with the number of training samples in the cluster to which the training sample belongs.
[0058] Among them, since anomaly detection is calculated based on data distribution density, the smaller the number of training samples in the cluster, the less accurate the duration anomaly of each training sample in the cluster. For example, if there is only one training sample in the cluster, the anomaly of the training sample has no reference value; at the same time, the smaller the number of training samples in the cluster, the greater the possibility that the concentrator state corresponding to the training sample is an abnormal state. Therefore, the selection probability of each training sample is calculated by combining the duration anomaly and the number of training samples in the cluster. The probability of selection Satisfies the relationship:
[0059] , For training samples The number of training samples in the cluster to which it belongs, is the total number of training samples, For training samples The duration of abnormality.
[0060] S24, selecting multiple batches of training samples according to the selection probability, and using mean square error loss to train the response time prediction model to obtain a trained response time prediction model.
[0061] Among them, a batch of training samples is randomly selected from all training samples, wherein the number of training samples in a batch is 16, 32 or 64, and the training samples with a greater selection probability are more effective for the training of the response time prediction model. In the process of selecting a batch of training samples, the training samples with a greater selection probability are more likely to be selected.
[0062] The batch of training samples is input into the response time prediction model to obtain the output result of each training sample, and the mean square error loss is calculated. The response time prediction model is updated using the gradient descent method to complete one training.
[0063] Select training samples from multiple training batches, and train the response time prediction model multiple times until the mean square error loss is less than the preset loss or the number of training times is greater than the preset number of times, to obtain a trained response time prediction model. The preset loss is 0.01, and the preset number of times is 500.
[0064] In this way, a trained response time prediction model is obtained. By selecting and excluding training samples with abnormal actual response time, it is ensured that the trained response time prediction model can accurately predict the standard response time of the heartbeat signal at the next moment.
[0065] The above-mentioned state monitoring method of the concentrator provided in the embodiment of the present application collects the real-time communication state of the concentrator at the current moment, and the number of electricity meters of each type of electricity consumption that needs to be read at the next moment, wherein the number of electricity meters of each type of electricity consumption that needs to be read directly determines the amount of data that needs to be transmitted at the next moment; the number of electricity meters of each type of electricity consumption that needs to be read at the next moment, and the real-time communication state are input into the trained response time prediction model to accurately predict the standard response time of the heartbeat signal at the next moment; a heartbeat signal is sent at the next moment, and if a confirmation signal is not received within the standard response time, it means that the state monitoring result of the concentrator is abnormal, thereby realizing real-time monitoring of the communication state of the concentrator.
[0066] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0067] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application.
Claims
1. A method for monitoring the status of a concentrator, characterized in that: Applied to the intelligent master station of the concentrator or any collector, the monitoring method includes: Input the real-time communication status of the concentrator and the number of meters of each type of electricity consumption that need to be read at the next moment into the trained response time prediction model to predict the standard response time of the heartbeat signal at the next moment, wherein the electricity consumption types include residential electricity consumption, industrial electricity consumption, commercial electricity consumption and public facility electricity consumption, and the real-time communication status includes the instantaneous bandwidth occupancy rate of each channel; Send a heartbeat signal at the next moment. If no confirmation signal is received after the interval standard response time, the status monitoring result of the concentrator is abnormal. Otherwise, the status monitoring result is normal. The response time prediction model is a BP neural network; The response time prediction model includes an input layer, multiple hidden layers and an output layer; the input layer is used to receive the real-time communication status and the number of electric meters of each type of electricity consumption; the multiple hidden layers are used to perform dimension transformation on the concatenated vector composed of the real-time communication status and the number of electric meters of each type of electricity consumption; the output layer is used to regress the output result of the last hidden layer to the standard response time; the input layer and each hidden layer include multiple neurons, and the activation function of each neuron is a nonlinear activation function; the output layer includes 1 neuron, and the activation function of the neuron is a ReLU function; The training method of the response time prediction model includes: collecting the communication status at a historical moment and the number of meters of each type of electricity consumption that need to be read at the next moment of the historical moment as training samples, and collecting the actual response time of each training sample; clustering the training samples to obtain multiple clusters, performing anomaly detection on the actual response time of the training samples in each cluster, and obtaining the time anomaly degree of each training sample; calculating the selection probability of each training sample, and the training sample The probability of selection Satisfies the relationship: , For training samples The number of training samples in the cluster to which it belongs, is the total number of training samples, For training samples ; selecting multiple batches of training samples according to the selection probability, and training the response time prediction model using mean square error loss to obtain a trained response time prediction model.
2. A concentrator status monitoring method as claimed in claim 1, characterized in that: The nonlinear activation function includes a sigmoid function and a tanh function.
3. A method for monitoring the status of a concentrator according to claim 1, characterized in that: The instantaneous bandwidth occupancy of a channel includes the bandwidth occupancy of the uplink and the bandwidth occupancy of the downlink.
4. A method for monitoring the status of a concentrator according to claim 1, characterized in that: The heartbeat signal includes preset information; in response to the time interval of receiving the confirmation signal being less than or equal to the standard response duration, and the preset information in the confirmation signal is the same as the preset information of the heartbeat signal, the status monitoring result of the concentrator is normal, otherwise, the status monitoring result is abnormal.
5. The method for monitoring the status of a concentrator according to claim 1, characterized in that: The clustering process for the training samples includes: The training samples are clustered based on a density clustering algorithm, where the density clustering algorithm is a DBSCAN algorithm or a HDBSCAN algorithm.
