Base station energy-saving optimization method in power wireless communication network

Abstract

The invention relates to the technical field of communication, and discloses a base station energy-saving optimization method in a power wireless communication network, which comprises the following steps of: acquiring base station energy equipment data; historical data of various energy devices of the base station are collected for data feature classification; calibrating an abnormal value, and analyzing the data type of the abnormal value; mapping the data type of the abnormal value with energy equipment to obtain targeted energy equipment, and predicting the abnormal rate of the targeted energy equipment based on the constructed grid model; obtaining an energy-saving scheme of each target energy device, generating an energy-saving optimization set, and selecting an optimal energy-saving optimization measure; the selected optimal energy-saving optimization measure is executed, meanwhile, corresponding early warning information is sent to the terminal, through the arrangement of the step S04 and the step S05, the grid model and the trained neural network model are combined and fused into fitness calculation of a natural enlightenment algorithm, data are fully utilized, and the fitness calculation efficiency is improved. And the influence of the energy-saving optimization measures on the energy equipment can be comprehensively and accurately known.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and more particularly to a method for optimizing energy saving of base stations in a power wireless communication network. Background Art

[0002] With the development of mobile communication technologies, the fifth-generation mobile communication has been developed today, and various performances of base stations have also been improved, so as to be able to adapt to different demand scenarios of high bandwidth, low latency, and large connection; however, while people enjoy higher and higher quality information services, the pressure on the base station side is also increasing continuously, and the energy consumed by it is also increasing rapidly, resulting in a corresponding increase in the annual expenditure of operators on base station energy consumption; most current base station energy-saving systems are based on traditional mathematical modeling to construct an analysis model, and then obtain energy-saving parameters by solving an optimization problem, and finally perform energy-saving operations according to the energy-saving configuration parameters. Currently, most base station energy-saving strategies are based on data such as base station load and traffic volume to perform threshold discrimination to determine whether to perform energy-saving operations on the target base station, such as base station sleep, carrier shutdown, multi-frequency coordination, etc.

[0003] The published document with the publication number CN112312531B discloses a base station energy-saving method and device, which determines energy-saving information according to the traffic loads of all cells in a target area; the energy-saving information includes the identifiers of energy-saving cells and energy-saving time periods; determines an energy-saving strategy according to the energy-saving information and the traffic loads of compensation cells; the energy-saving strategy is used to indicate the energy-saving method of the energy-saving cells during the energy-saving time periods, and the energy-saving strategy includes the identifiers of the energy-saving cells, the energy-saving time periods, the traffic loads of the energy-saving cells, and the identifiers of the compensation cells; the compensation cells are used to carry the traffic loads of the energy-saving cells; optimizes the energy-saving cells according to the energy-saving strategy, so that the energy-saving cells blind-compensate the cells in the co-covered cell set or the neighboring cells of the energy-saving cells, so as to provide better network service quality while saving energy for the base station.

[0004] However, in the existing methods for saving energy of base stations, the base stations are all considered and optimized for energy saving as primary units, and adjustment and coordination are performed among each base station to achieve the purpose of overall energy-saving optimization. However, for individual base stations that cannot be coordinately adjusted, adaptive adjustment is required to achieve the purpose of energy saving; base station equipment is the main source of base station energy consumption, and the energy consumption situation of base station equipment will change over time. Therefore, it is necessary to regularly monitor the energy consumption data of the equipment and perform necessary maintenance work to ensure that the energy efficiency of the equipment is always in the optimal state, so as to perform adaptive adjustment of the base station and then achieve the purpose of energy-saving optimization. Summary of the Invention

[0005] To overcome the above defects of the prior art, the present invention provides an energy-saving optimization method for base stations in a power wireless communication network to solve the problems existing in the above background art.

[0006] The present invention provides the following technical solutions: An energy-saving optimization method for base stations in a power wireless communication network, comprising the following steps:

[0007] Step S01: Collect data of base station energy equipment: Collect energy consumption data and equipment operation parameter data of n energy equipment in the base station for each time period;

[0008] Step S02: Collect historical data of various types of energy equipment in the base station and classify the data characteristics, and classify the data into fault characteristic data, aging characteristic data, and normal fluctuation characteristic data;

[0009] Step S03: Calibrate the outliers in the base station energy equipment data collected in Step S01, and analyze the data types of the outliers;

[0010] Step S04: Map the data types of the outliers to the energy equipment to obtain the targeted energy equipment, and predict the outlier rate of the targeted energy equipment based on the constructed grid model;

[0011] Step S05: Obtain the energy-saving solutions for each targeted energy equipment, generate an energy-saving optimization set, and select the best energy-saving optimization measures based on the outlier rate of the energy equipment predicted in Step S04;

[0012] Step S06: Execute the selected best energy-saving optimization measures, and at the same time send corresponding warning information to the terminal based on the outlier rate of the targeted energy equipment predicted in Step S04.

[0013] Preferably, the energy consumption data is the type of energy consumed by the energy equipment and the energy consumption, and the equipment operation parameter data includes antenna transmission power, operating frequency, receiving sensitivity, and power supply output power;

[0014] The historical data includes historical energy consumption data and equipment operation parameter data. The various types of energy equipment in the base station are specifically three types: faulty energy equipment, aging energy equipment, and normal energy equipment. Collect the historical energy consumption data and equipment operation parameter data of the faulty energy equipment, aging energy equipment, and normal energy equipment;

[0015] The data types include fault characteristic data types, aging characteristic data types, and normal fluctuation characteristic data types;

[0016] The outlier rate includes the failure rate and the aging rate.

[0017] Preferably, the data classification in Step S02 includes the following steps:

[0018] Step S11: Extract the data features of the historical data of various energy devices of the base station, and use the data features as sample points. Randomly select one sample point from the historical data features of the faulty energy devices, aging energy devices, and normal energy devices as the initial clustering centers, and label them as z1, z2, and z3 in sequence;

[0019] Step S12: Denote the sample points that are not used as clustering centers as calculation points, and label them as j1, j2, j3, …, j m-3 ; Calculate the distance D ba from each calculation point to each clustering center in sequence. The calculation formula is expressed as: where D ba is the distance from the b-th calculation point j b to the a-th clustering center z a , where b = 1, 2, 3, …, m - 3; a = 1, 2, 3; R is the dimension of the sample point, u = 1, 2, 3, …, R; j bu is the coordinate of the b-th calculation point j b in the u-th dimension, and z au is the coordinate of the a-th clustering center z a in the u-th dimension; m is the total number of data, and each data corresponds to a sample point;

[0020] Step S13: Establish three corresponding clusters based on the three clustering centers;

[0021] Step S14: Compare the distances from the calculation point j b to each clustering center, and assign the calculation point to the cluster corresponding to the nearest clustering center;

[0022] Step S15: Let b = b + 1, and jump back to Step S14;

[0023] Step S16: Repeat Steps S14 to S15 until the loop ends when b = m - 3, and assign the m - 3 calculation points to the corresponding clusters;

[0024] Step S17: Recalculate the coordinates of the new clustering center of each cluster. The calculation formula is expressed as: where z′ c is the coordinate of the new clustering center of the c-th cluster, c = 1, 2, 3, y i is the coordinate of the i-th calculation point in the c-th cluster, i = 1, 2, 3, …, I, and I is the total number of calculation points in the c-th cluster, y i =(y i1 ,y i2 ,y i3 ,...,y iR );

[0025] Step S18: Repeat steps S12 to S17 until the newly recalculated cluster center coordinates for each cluster in step S17 are the same as those calculated in the previous cycle, at which point the loop ends; obtain the clusters corresponding to the 3 clusters and the corresponding sample points, and use the data type of the cluster center of each cluster as the data cluster type. The sample points in the cluster are all data features of the corresponding data type.

[0026] Preferably, the specific method for analyzing the data type of outliers in step S03 is as follows:

[0027] Annotate the real-time energy consumption data of n energy devices of the base station, and annotate the energy consumption of the f-th energy device at the t-th moment as θ ft , where f = 1, 2, 3, …, n, t = 1, 2, 3, …, T, and T is the total number of moments;

[0028] Obtain the average energy consumption of n energy devices. The calculation formula is expressed as: where, P f is the average energy consumption of the f-th energy device;

[0029] Calibrate the values higher than the average energy consumption as outliers. The outliers of each energy device form an outlier set U. Then the outlier set of the f-th energy device is expressed as: U f = {uf1, uf2, uf3,..., uf q}, where U f is the outlier set of the f-th energy device, and uf q is the q-th outlier data of the f-th energy device; obtain the outlier sets of n energy devices;

[0030] Calculate the distance from the data features of each outlier in the outlier set of each energy device to the new cluster center point of each cluster in step S18, and perform an ascending order from small to large. Allocate each outlier data to the cluster corresponding to the cluster center point with the closest distance, that is, the cluster corresponding to the cluster center point with the smallest distance value. The data type corresponding to this cluster is the data type of the outlier data allocated to this cluster;

[0031] The calculation formula is: where, Df qc is the distance from the q-th outlier data in the outlier set of the f-th energy device to the new cluster center point of the c-th cluster, fq u is the coordinate of the q-th outlier data in the outlier set of the f-th energy device in the u-th dimension, z′ cu is the coordinate of the new cluster center point z′ c of the c-th cluster in the u-th dimension, R is the dimension in step S12, and u = 1, 2, 3, …, R;

[0032] Obtain the data types of each outlier in each outlier set of energy devices.

[0033] Preferably, the specific method of mapping the data type of the outlier to the energy device in step S04 is as follows:

[0034] Establish a one-to-one mapping relationship between each outlier set of energy devices and the corresponding energy device, calculate the proportion of the data types of each outlier in each outlier set of energy devices, add the proportion of the fault feature data type and the proportion of the aging feature data type. If it exceeds the set threshold, the energy device mapped by the outlier set is recorded as the target energy device;

[0035] The formula is expressed as: Among them, Zf1 is the proportion of the fault feature data type in the f-th outlier set of energy devices, Zf2 is the proportion of the aging feature data type in the f-th outlier set of energy devices, Zf3 is the proportion of the normal fluctuation feature data type in the f-th outlier set of energy devices, and Qf is the number of all outliers in the f-th outlier set of energy devices;

[0036] If Zf1 + Zf2 ≥ YU, the energy device mapped by the outlier set is recorded as the target energy device, where YU is the set threshold.

[0037] Preferably, the method for establishing the grid model is as follows:

[0038] According to the physical structure and components of each target energy device, discretize the target energy device into a grid structure. The grid structure contains several grid cells, and each grid cell represents a physical structure or component of the target energy device; set a discrete state for each grid cell, such as a normal state, an aging state, and a fault state; the discrete state is represented by binary coding, and the state of the grid cell can be defined as {0: normal state; 1: aging state; 2: fault state}; define the state transition rule of the w-th grid cell, and represent the state transition rule with a Boolean function; this grid structure is the grid model;

[0039] The state transition rule is: if itself or any adjacent grid cell is in the fault state, then transfer to the fault state; otherwise, if itself or any adjacent grid cell is in the aging state, then transfer to the aging state.

[0040] Preferably, the specific method for predicting the outlier rate of the target energy device based on the constructed grid model is as follows:

[0041] For each target energy device, update the state of each of its grid cells until the current moment to obtain a real-time grid model;

[0042] Based on the overall grid state of the real-time grid model, update the overall grid state at the next moment, obtain the failure rate and aging rate in the overall grid state at the next moment, and get the anomaly rate of the targeted energy equipment at the next moment;

[0043] The failure rate is the proportion of grid cells in the failure state in the overall grid state at the next moment; the aging rate is the proportion of grid cells in the aging state in the overall grid state at the next moment; add the value obtained by multiplying the failure rate by the corresponding weight factor and the value obtained by multiplying the aging rate by the corresponding weight factor to get the anomaly rate; thus, obtain the anomaly rate of each targeted energy equipment.

[0044] Preferably, the way of state update is:

[0045] For the e-th time step, traverse all grid cells. For the w-th grid cell, obtain the states of itself and its adjacent grid cells at the (e - 1)-th time step; calculate the state of the w-th grid cell at the e-th time step according to the state transition rule, and update the state of the w-th grid cell. Repeat and record the states of all grid cells at the e-th time step, which is the overall grid state, to form a grid model; the time step is the time difference between the current moment and the previous moment or the next moment.

[0046] Preferably, the best energy-saving optimization measure is obtained by using an improved nature-inspired algorithm, specifically:

[0047] Step S21: Encode the energy-saving optimization set, and the encoding is O. O is the chromosome. Obtain the chromosome and construct the initial population A = {O1, O2, O3,..., O k};

[0048] Step S22: Determine the fitness function;

[0049] Step S23: Conduct natural selection on the chromosomes in the population;

[0050] Step S24: Conduct crossover recombination on the chromosomes in the population;

[0051] Step S25: Conduct mutation on the chromosomes in the population;

[0052] Step S26: Obtain the new population. Preset the population algebra as L and the fitness threshold as Q. L is an integer greater than 0, and Q is a real number greater than 0; loop from step S23 to step S25 until the generation corresponding to the new population is L or there is a chromosome in the new population whose corresponding fitness is greater than or equal to the fitness threshold Q. The loop ends, and obtain the energy-saving optimization set corresponding to the chromosome with the maximum fitness in the new population as the best energy-saving optimization set. The energy-saving optimization measures in the best energy-saving optimization set are the best energy-saving optimization measures.

[0053] Preferably, the expression of the fitness function is: λ r =-YZ r , where λ r is the fitness corresponding to the r-th chromosome, and YZ r is the abnormal rate suppression value of the energy-saving optimization set corresponding to the r-th chromosome; r = 1, 2, 3, …, k;

[0054] The way to obtain the abnormal rate suppression value is as follows:

[0055] Take the energy-saving optimization measures in the energy-saving optimization set as input data, input them into the energy-saving optimization measure prediction model, obtain the abnormal rates of each energy device corresponding to the energy-saving optimization measures in the energy-saving optimization set as output, perform weighted average on the output energy device abnormal rates to obtain the average abnormal rate, and subtract the predicted average value of the energy device abnormal rate to obtain the abnormal rate suppression value;

[0056] The way to construct the energy-saving optimization measure prediction model is as follows:

[0057] Take the energy-saving optimization measures in the energy-saving optimization set, the energy consumption data of the corresponding energy devices after energy-saving optimization, and the device operation parameter data as analysis data. Pre-collect d groups of analysis data, where d is an integer greater than 1. Convert the analysis data and the corresponding energy device abnormal rates into a corresponding set of feature vectors; each group of analysis data is the energy-saving optimization measures of an energy device and the corresponding energy consumption data of the energy device after energy-saving optimization and the device operation parameter data;

[0058] Take each group of feature vectors as the input of the energy-saving optimization measure prediction model. The energy-saving optimization measure prediction model takes a set of predicted energy device abnormal rates corresponding to each group of analysis data as output, takes the actual energy device abnormal rate corresponding to each group of analysis data as the prediction target, and the actual energy device abnormal rate is the energy device abnormal rate pre-collected corresponding to the analysis data; take minimizing the sum of the prediction errors of all analysis data as the training target; the formula for the prediction error is expressed as: ε p =σ p -μ p , where ε p is the prediction error, p is the group number of the feature vector corresponding to the analysis data, σ p is the predicted energy device abnormal rate corresponding to the p-th group of analysis data, μ p is the actual energy device abnormal rate corresponding to the p-th group of analysis data. Train the energy-saving optimization measure prediction model until the sum of the prediction errors reaches convergence and then stop training. The technical effects and advantages of the present invention:

[0059] By providing step S04 and step S05, the present invention is conducive to obtaining the predicted failure rate and aging rate of energy equipment by constructing a grid model, thereby predicting the abnormal rate of the equipment, and adopting an improved natural inspiration algorithm to obtain the best energy-saving optimization measures. Combining the grid model with the trained neural network model and integrating it into the fitness calculation of the natural inspiration algorithm can give full play to the non-linear and multi-modal data modeling capabilities of the neural network, effectively capture complex data relationships; according to the parallel computing processing mechanism, improve the computing efficiency, and evaluate the abnormal rate suppression effect of energy equipment through deep learning technology, make full use of data, capture complex data relationships through deep learning technology, can comprehensively and accurately understand the impact of energy-saving optimization measures on energy equipment, ensure that the energy-saving optimization measures are the best energy-saving methods, based on ensuring energy equipment, conduct energy conservation, select energy-saving optimization measures that can maximize the guarantee of the use of energy equipment, thereby promoting the sustainable development of energy equipment, and then through the adaptive adjustment of the base station energy equipment, make the energy equipment always in the optimal state, and then ensure that the base station is in the best energy-saving state as a whole. Description of the Drawings

[0060] Figure 1 It is a flowchart of the base station energy-saving optimization method in the power wireless communication network of the present invention. Detailed Embodiments

[0061] Next, the technical solutions in the present invention will be clearly and completely described in conjunction with the drawings in the present invention. In addition, the forms of each structure described in the following embodiments are merely examples, and a base station energy-saving optimization method related to the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0062] As Figure 1 shown, the present invention provides a base station energy-saving optimization method in a power wireless communication network, including the following steps:

[0063] Step S01: Collect data of base station energy equipment: Collect the energy consumption data and equipment operation parameter data of n energy equipment in the base station in each time period; the energy consumption data is the types of energy consumed by the energy equipment and the energy consumption, and the equipment operation parameter data includes, but is not limited to, data describing the operation state of the equipment such as antenna transmission power, working frequency, receiving sensitivity, and power supply output power.

[0064] Step S02: Collect historical data of various energy devices in the base station for data feature classification, and classify the data into fault feature data, aging feature data, and normal fluctuation feature data; the historical data includes historical energy consumption data and device operation parameter data, and the various energy devices in the base station are specifically three types: faulty energy devices, aging energy devices, and normal energy devices. Collect the historical energy consumption data and device operation parameter data of faulty energy devices, aging energy devices, and normal energy devices.

[0065] Step S03: Calibrate the outliers in the base station energy device data collected in Step S01, and analyze the data types of the outliers; the data types include fault feature data type, aging feature data type, and normal fluctuation feature data type.

[0066] Step S04: Map the data types of the outliers to the energy devices to obtain the targeted energy devices, and predict the outlier rates of the targeted energy devices based on the constructed grid model; the outlier rates include failure rate and aging rate; by obtaining the prediction of the failure rate and aging rate of the energy devices, the future performance of the energy devices can be obtained, and then different energy-saving measures can be collected according to different energy device conditions. Before replacing the energy devices, delay the aging of the energy devices, reduce the failure rate of the energy devices, and prevent the increase in the overall energy consumption of the base station caused by the failure or aging of the energy devices.

[0067] Step S05: Obtain the energy-saving solutions for each targeted energy device, generate an energy-saving optimization set, and select the best energy-saving optimization measure based on the outlier rate of the energy devices predicted in Step S04; the energy-saving solutions for each energy device can be obtained by those skilled in the art by collecting historical energy-saving measures and combining technical knowledge in the professional field. The energy-saving solutions are all solutions that meet the user needs in the area where the base station is located. Different energy devices have different energy-saving measures. For energy devices that may have failures or aging in the future, corresponding energy-saving measures are collected, which can reduce the increase in energy consumption caused by device failures or aging while ensuring the normal use of the devices, and ensure that other energy devices are not affected, so as not to affect the normal network communication services of the base station.

[0068] Step S06: Execute the selected best energy-saving optimization measure, and at the same time send corresponding warning information to the terminal based on the outlier rate of the targeted energy devices predicted in Step S04 to prompt the technical personnel to arrange repairs or maintenance to ensure the normal operation of the base station.

[0069] In this embodiment, it should be specifically noted that the data classification in Step S02 includes the following steps:

[0070] Step S11: Extract the data features of the historical data of various energy devices of the base station, and use the data features as sample points. Randomly select one sample point from the historical data features of faulty energy devices, aging energy devices, and normal energy devices as the initial clustering centers, and label them as z1, z2, and z3 in sequence;

[0071] Step S12: Denote the sample points that are not used as clustering centers as calculation points, and label them as j1, j2, j3, …, j m-3 ; Calculate the distance D ba from each calculation point to each clustering center in sequence. The calculation formula is expressed as: where D ba is the distance from the b-th calculation point j b to the a-th clustering center z a , where b = 1, 2, 3, …, m - 3; a = 1, 2, 3; R is the dimension of the sample point, u = 1, 2, 3, …, R; j bu is the coordinate of the b-th calculation point j b in the u-th dimension, and z au is the coordinate of the a-th clustering center z a in the u-th dimension; m is the total number of data, and each data corresponds to a sample point;

[0072] Step S13: Establish three corresponding clusters based on the three clustering centers;

[0073] Step S14: Compare the distances from the calculation point j b to each clustering center, and assign the calculation point to the cluster corresponding to the nearest clustering center;

[0074] Step S15: Let b = b + 1, and jump back to Step S14;

[0075] Step S16: Repeat Steps S14 to S15 until the loop ends when b = m - 3, and assign the m - 3 calculation points to the corresponding clusters;

[0076] Step S17: Recalculate the coordinates of the new clustering center of each cluster. The calculation formula is expressed as: where z′ c is the coordinate of the new clustering center of the c-th cluster, c = 1, 2, 3, y i is the coordinate of the i-th calculation point in the c-th cluster, i = 1, 2, 3, …, I, and I is the total number of calculation points in the c-th cluster, y i =(y i1 ,y i2 ,y i3 ,...,y iR );

[0077] Step S18: Repeat steps S12 to S17 until the newly recalculated cluster center coordinates for each cluster in step S17 are the same as those calculated in the previous cycle, at which point the loop ends; Obtain the clusters corresponding to 3 clusters and the corresponding sample points, and use the data type of the cluster center of each cluster as the data cluster type. The sample points in the cluster are all data features of the corresponding data type. Exemplarily, if the data type of the cluster center in the first cluster is fault feature data, then the data cluster type is the fault type, and the sample points in the cluster are all fault feature data, that is, the energy consumption data features that occur when the energy device fails; Aging data is the energy consumption data features that occur when the energy device ages;

[0078] There will be a phenomenon of data interspersion among faulty energy devices, aging energy devices, and normal energy devices. That is, there will be aging data features in faulty energy devices, fault data features in aging energy devices, and both fault data features and aging data features in normal energy devices. Therefore, if only classifying data features according to faulty energy devices, aging energy devices, and normal energy devices, the accuracy is not high, and the characteristics of the data are not analyzed deeply enough. Using a clustering algorithm to cluster the data can better capture the data characteristics, discover the hidden information in the data, help understand the hierarchical relationship between the data, and provide useful help for subsequent data analysis and decision support.

[0079] In this embodiment, it should be specifically noted that the specific method for analyzing the data type of outliers in step S03 is as follows:

[0080] Label the real-time energy consumption data of n energy devices in the base station. Label the energy consumption of the f-th energy device at the t-th moment as θ ft , f = 1, 2, 3,..., n, t = 1, 2, 3,..., T, where T is the total number of moments;

[0081] Obtain the average energy consumption of n energy devices. The calculation formula is expressed as: where, P f is the average energy consumption of the f-th energy device;

[0082] Calibrate the values higher than the average energy consumption as outliers. The outliers of each energy device form an outlier set U. Then the outlier set of the f-th energy device is expressed as: U f ={uf1, uf2, uf3,..., uf q}, where, U f is the outlier set of the f-th energy device, and uf q is the q-th outlier data in the f-th energy device; Obtain the outlier sets of n energy devices;

[0083] Calculate the distances from the data characteristics of each outlier in the outlier set of each energy device to the new cluster center points in step S18, and arrange them in ascending order from smallest to largest. Assign each outlier data to the cluster corresponding to the nearest cluster center point, that is, the cluster corresponding to the cluster center point with the smallest distance value. The data type corresponding to this cluster is the data type of the outlier data assigned to this cluster;

[0084] The calculation formula is: where Df qc is the distance from the q-th outlier data in the outlier set of the f-th energy device to the new cluster center point of the c-th cluster, fq u is the coordinate of the q-th outlier data in the outlier set of the f-th energy device in the u-th dimension, z′ cu is the coordinate of the new cluster center point z′ of the c-th cluster c in the u-th dimension, R is the dimension in step S12, and u = 1, 2, 3,..., R;

[0085] Obtain the data types of each outlier in the outlier set of each energy device, and then it can be obtained what kind of situation of the energy device causes the corresponding energy consumption data of the outlier. If the data type of the outlier is fault characteristic data, that is, the abnormality of the energy consumption data is caused by equipment failure. Therefore, taking corresponding energy-saving optimization measures to reduce the fault of the equipment can reduce the energy consumption generated by the equipment failure, and further achieve the purpose of energy-saving for the base station while extending the service life of the energy device.

[0086] In this embodiment, it should be specifically noted that the specific method of mapping the data type of the outlier to the energy device in step S04 is:

[0087] Establish a one-to-one mapping relationship between each outlier set of each energy device and the corresponding energy device. Calculate the proportion of each outlier data type in each outlier set of each energy device. Add the proportion of the fault characteristic data type and the proportion of the aging characteristic data type. If it exceeds the set threshold, mark the energy device mapped by this outlier set as the target energy device;

[0088] The formula is expressed as: where Zf1 is the proportion of the fault characteristic data type in the outlier set of the f-th energy device, Zf2 is the proportion of the aging characteristic data type in the outlier set of the f-th energy device, Zf3 is the proportion of the normal fluctuation characteristic data type in the outlier set of the f-th energy device, and Qf is the number of all outliers in the outlier set of the f-th energy device;

[0089] If Zf1 + Zf2 ≥ YU, mark the energy device mapped by this outlier set as the target energy device, where YU is the set threshold, and YU ≤ 50%.

[0090] In this embodiment, it should be specifically noted that the method for establishing the grid model includes:

[0091] According to the physical structure and components of each target energy device, the target energy device is discretized into a grid structure, which contains several grid cells, and each grid cell represents a physical structure or component of the target energy device; a discrete state is set for each grid cell, such as a normal state, an aging state, and a fault state; the discrete state is represented by a binary code, and the state of the grid cell can be defined as {0: normal state; 1: aging state; 2: fault state}; the state transition rule of the w-th grid cell is defined and represented by a Boolean function; this grid structure is the grid model;

[0092] The state transition rule is: if itself or any adjacent grid cell is in the fault state, then it transfers to the fault state; otherwise, if itself or any adjacent grid cell is in the aging state, then it transfers to the aging state;

[0093] The historical data of the base station energy device is used to verify the accuracy of the state transition rule, and according to the verification result, the rule is adjusted and optimized to better reflect the actual aging and fault processes;

[0094] An initial discrete state is assigned to each grid cell, and the initial discrete state is set based on the historical data of the base station energy device; it can reflect the operating state of the energy device at different time points; by analyzing these data, the aging degree of the device is evaluated, and the operating state of the device is judged, so as to assign a reasonable initial discrete state value to the grid cell.

[0095] In this embodiment, it should be specifically noted that the specific method for predicting the abnormality rate of the target energy device based on the constructed grid model is as follows:

[0096] For each target energy device, the state of each of its grid cells is updated until the current moment to obtain a real-time grid model;

[0097] Based on the overall grid state of the real-time grid model, the overall grid state at the next moment is updated, and the failure rate and aging rate in the overall grid state at the next moment are obtained to get the abnormality rate of the target energy device at the next moment;

[0098] The failure rate is the proportion of grid cells in a failed state in the overall grid state at the next moment; the aging rate is the proportion of grid cells in an aging state in the overall grid state at the next moment; the sum of the value obtained by multiplying the failure rate by the corresponding weight factor and the value obtained by multiplying the aging rate by the corresponding weight factor is the anomaly rate; both the failure rate corresponding weight factor and the aging rate corresponding weight factor satisfy being greater than 0 and less than 1, and the failure rate corresponding weight factor is greater than the aging rate corresponding weight factor. The specific values of the weight factors are adaptively set by those skilled in the art according to the impact degree of the energy device on the base station, and this embodiment does not make specific limitations on this.

[0099] Thus, the anomaly rate of each targeted energy device is obtained.

[0100] The way of state update is as follows:

[0101] For the e-th time step, all grid cells are traversed. For the w-th grid cell, the states of itself and its adjacent grid cells at the (e - 1)-th time step are obtained; according to the state transition rule, the state of the w-th grid cell at the e-th time step is calculated, and the state of the w-th grid cell is updated. The states of all grid cells at the e-th time step are repeated and recorded, which is the overall grid state, forming a grid model; the time step is the time difference between the current moment and the previous moment or the next moment.

[0102] In this embodiment, it should be specifically noted that the energy-saving scheme includes multiple different energy-saving measures, and B v is the energy-saving scheme for the v-th targeted energy device. If there are a total of V targeted energy devices, one energy-saving measure is selected from the energy-saving schemes of the V targeted energy devices for permutation and combination to form several energy-saving optimization sets, and the best energy-saving optimization measure is selected from them.

[0103] The selection of the best energy-saving optimization measure is obtained by using an improved natural inspiration algorithm. The natural inspiration algorithm can be any one of natural inspiration algorithms such as genetic algorithm, particle swarm algorithm, bat algorithm, etc. In this embodiment, the genetic algorithm is selected; specifically:

[0104] Step S21: Encode the energy-saving optimization set. The encoding is O, and O is the chromosome. The chromosome is obtained, and an initial population A = {O1, O2, O3,..., O k} is constructed.

[0105] Step S22: Determine the fitness function.

[0106] Step S23: Perform natural selection on the chromosomes in the population.

[0107] Step S24: Perform crossover recombination on the chromosomes in the population.

[0108] Step S25: Mutate the chromosomes in the population;

[0109] Step S26: Obtain a new population. The preset population generation number is L, and the fitness threshold is Q. L is an integer greater than 0, and Q is a real number greater than 0. Loop from Step S23 to Step S25 until the generation number corresponding to the new population is L or there is a chromosome in the new population whose corresponding fitness is greater than or equal to the fitness threshold Q. Then the loop ends, and the energy-saving optimization set corresponding to the chromosome with the maximum fitness in the new population is obtained as the optimal energy-saving optimization set, and the energy-saving optimization measures in the optimal energy-saving optimization set are the optimal energy-saving optimization measures. Exemplarily, if the preset population generation number is 1, then natural selection, crossover recombination, and mutation are performed on the chromosomes in the initial population to obtain a new population. At this time, the generation number corresponding to the new population is 1, so the loop ends;

[0110] The fitness threshold Q is preset by those skilled in the art according to the algorithm accuracy. The population generation number L is obtained by those skilled in the art through multiple uses of the genetic algorithm under multiple groups of different energy-saving optimization sets and corresponding energy consumption data and equipment operation parameter data. In each use of the genetic algorithm, when there is a chromosome in the new population whose corresponding fitness is greater than or equal to the fitness threshold Q, the loop ends, and the generation number corresponding to the new population is obtained; the largest generation number among multiple generation numbers is used as the population generation number L.

[0111] In this embodiment, it should be specifically noted that the expression of the fitness function is: λ r =-YZ r , where λ r is the fitness corresponding to the r-th chromosome, and YZ r is the abnormal rate suppression value of the energy-saving optimization set corresponding to the r-th chromosome; r = 1, 2, 3,..., k;

[0112] The natural selection is performed by combining the elite method and the roulette method; among them, the elite method generates F1 offspring chromosomes. For a population with a capacity of k, the fitnesses corresponding to the k chromosomes are arranged from largest to smallest, and each of the top F1 chromosomes generates an offspring chromosome; the roulette method generates F2 offspring chromosomes, that is, the k chromosomes generate F2 offspring chromosomes according to the corresponding roulette probabilities; F1 + F2 = k to keep the offspring population capacity k unchanged and the population generation number increasing;

[0113] The expression of the roulette probability is: where ζ r is the roulette probability corresponding to the r-th chromosome;

[0114] In step S14, randomly select E chromosomes from the population for crossover recombination to obtain E new chromosomes. The PMX method is used for crossover recombination, which is a prior art means and will not be elaborated in detail in this embodiment. After chromosome crossover recombination, calculate the fitness of the E new chromosomes, sort the fitness of the E new chromosomes and the fitness of the E chromosomes from large to small to generate a sorting table, and replace the E chromosomes in the population that have undergone crossover recombination with the E new chromosomes in the sorting table in ascending order. In this embodiment, it is preferred that E = 0.7k. If the calculated E is not an integer, round E up to ensure that the calculated E is an integer.

[0115] In step S15, a preset mutation probability is H. Mutate k chromosomes in the population according to the mutation probability. The mutation method is to randomly select the positions of two genes in the chromosome and exchange the values of the two genes. In this embodiment, it is preferred that H = 0.02, and the mutation probability is preset by those skilled in the art according to the algorithm efficiency and algorithm accuracy.

[0116] In this embodiment, it should be specifically noted that the acquisition method of the abnormal rate suppression value is as follows:

[0117] Take the energy-saving optimization measures in the energy-saving optimization set as input data, input them into the energy-saving optimization measure prediction model, obtain the abnormal rates of each energy device corresponding to the energy-saving optimization measures in the energy-saving optimization set as output, perform weighted averaging on the output abnormal rates of the energy devices to obtain the average abnormal rate, and subtract the predicted average abnormal rate of the energy devices to obtain the abnormal rate suppression value. The abnormal rate suppression value can be positive or negative. When it is negative, it indicates that the abnormal rate of the energy device can be suppressed, and when it is positive, it indicates that the abnormal rate of the energy device cannot be suppressed.

[0118] The construction method of the energy-saving optimization measure prediction model is as follows:

[0119] Take the energy-saving optimization measures in the energy-saving optimization set, the energy consumption data of the corresponding energy devices after energy-saving optimization, and the device operation parameter data as analysis data. Pre-collect d sets of analysis data, where d is an integer greater than 1, and convert the analysis data and the corresponding abnormal rates of the energy devices into a corresponding set of feature vectors. Each set of analysis data is the energy-saving optimization measure of an energy device and the corresponding energy consumption data of the energy device after energy-saving optimization and the device operation parameter data.

[0120] The abnormal rate of the energy device corresponding to the analysis data can be calculated through the grid model constructed in step S04. Specifically, calculate the mean failure rate and mean aging rate of all current time steps, and add the value obtained by multiplying the mean failure rate by the corresponding weight factor and the value obtained by multiplying the mean aging rate by the corresponding weight factor. Collect d sets of analysis data, and under the conditions of each set of analysis data, comprehensively analyze the influence of different energy-saving optimization measures on energy devices.

[0121] Use each group of feature vectors as the input of the energy-saving optimization measure prediction model. The energy-saving optimization measure prediction model outputs a predicted energy equipment abnormality rate corresponding to each group of analysis data, and uses the actual energy equipment abnormality rate corresponding to each group of analysis data as the prediction target. The actual energy equipment abnormality rate is the energy equipment abnormality rate collected in advance corresponding to the analysis data. The training objective is to minimize the sum of the prediction errors of all analysis data. The formula for the prediction error is expressed as: ε p =σ p -μ p , where ε p is the prediction error, p is the group number of the feature vector corresponding to the analysis data, σ p is the predicted energy equipment abnormality rate corresponding to the p-th group of analysis data, and μ p is the actual energy equipment abnormality rate corresponding to the p-th group of analysis data. Train the energy-saving optimization measure prediction model until the sum of the prediction errors converges and then stop training;

[0122] The energy-saving optimization measure prediction model is specifically a deep neural network model, which includes an input layer, a hidden layer, and an output layer. Each hidden layer includes multiple neurons, and there are connections between each neuron and the neurons in the next layer. The connections contain weights, which determine the importance and influence of data transmission in the neural network. An activation function is applied to each neuron between the hidden layer and the output layer. The activation function introduces non-linearity and allows the network to learn more complex patterns and features.

[0123] In this embodiment, it should be specifically noted that the specific method for the step S06 to send the corresponding warning information to the terminal based on the predicted target energy equipment abnormality rate in the step S04 is as follows:

[0124] When the target energy equipment abnormality rate exceeds the warning threshold, send a corresponding warning prompt to the terminal. The warning prompt includes a warning sound and warning display content. The warning display content includes the target energy equipment and the corresponding failure rate and aging rate. Those skilled in the art perform repairs or maintenance based on the values of the failure rate and aging rate. For example, when the failure rate is 60% and the aging rate is 5%, it indicates that the degree of equipment aging is not serious, the equipment may be relatively new, but the failure rate is high, indicating that the equipment fails not due to aging but due to other failure reasons. At this time, timely repairs are required to ensure the normal operation of the equipment and reduce the increase in energy consumption caused by equipment failures;

[0125] When the target energy equipment abnormality rate does not exceed the warning threshold, only send the warning display content to the terminal without generating a warning prompt sound.

[0126] In this embodiment, it should be specifically noted that the main difference between this embodiment and the prior art is that this embodiment includes steps S04 and S05. By constructing a grid model, the predicted failure rate and aging rate of energy equipment are obtained, so as to predict the abnormal rate of the equipment. And an improved nature-inspired algorithm is used to obtain the best energy-saving optimization measures. Combining the grid model with the trained neural network model and integrating it into the fitness calculation of the nature-inspired algorithm can give full play to the non-linear and multi-modal data modeling capabilities of the neural network, effectively capture complex data relationships. According to the parallel computing processing mechanism, the computing efficiency is improved, and the effect of suppressing the abnormal rate of energy equipment is evaluated through deep learning technology. By making full use of data and capturing complex data relationships through deep learning technology, it is possible to comprehensively and accurately understand the impact of energy-saving optimization measures on energy equipment, ensure that the energy-saving optimization measures are the best energy-saving methods, and based on ensuring energy equipment, energy conservation is carried out. The energy-saving optimization measures that can ensure the use of energy equipment to the greatest extent are selected, so as to promote the sustainable development of energy equipment. Then, through the adaptive adjustment of the base station energy equipment, the energy equipment is always in the optimal state, and thus the overall base station is ensured to be in the best energy-saving state.

[0127] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

[0128] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or replacements, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for optimizing the energy saving of base stations in a power wireless communication network, characterized in that: Including the following steps: Step S01: Collect data of base station energy equipment: Collect energy consumption data and equipment operation parameter data of n energy equipment in the base station for each time period; Step S02: Collect historical data of various types of energy equipment in the base station and classify data features, and classify the data into fault feature data, aging feature data, and normal fluctuation feature data; Step S03: Calibrate outliers in the base station energy equipment data collected in Step S01, and analyze the data types of the outliers; Step S04: Map the data types of the outliers to the energy equipment to obtain the targeted energy equipment, and predict the outlier rate of the targeted energy equipment based on the constructed grid model; Step S05: Obtain the energy-saving solutions for each targeted energy equipment, generate an energy-saving optimization set, and select the best energy-saving optimization measures based on the outlier rate of the energy equipment predicted in Step S04; Step S06: Execute the selected best energy-saving optimization measures, and at the same time send corresponding warning information to the terminal based on the outlier rate of the targeted energy equipment predicted in Step S04.

2. The energy-saving optimization method for a base station in a power wireless communication network according to claim 1, characterized in that: The energy consumption data is the type of energy consumed by the energy equipment and the energy consumption, and the equipment operation parameter data includes antenna transmit power, operating frequency, receiving sensitivity, and power supply output power; The historical data includes historical energy consumption data and equipment operation parameter data. The specific types of various energy equipment in the base station are three types: faulty energy equipment, aging energy equipment, and normal energy equipment. Collect the historical energy consumption data and equipment operation parameter data of faulty energy equipment, aging energy equipment, and normal energy equipment; The data types include fault feature data type, aging feature data type, and normal fluctuation feature data type; The outlier rate includes failure rate and aging rate.

3. A base station energy-saving optimization method in a power wireless communication network according to claim 2, characterized in that: The data classification in Step S02 includes the following steps: Step S11: Extract the data features of the historical data of various types of energy equipment in the base station, and use the data features as sample points. Randomly select one sample point from the data features of the historical data of faulty energy equipment, aging energy equipment, and normal energy equipment as the initial clustering center, and mark them as z1, z2, and z3 in turn; Step S12: Denote the sample points that are not cluster centers as calculation points, and label them as j1, j2, j3, …, j m-3 ; Calculate the distance D from each calculation point to each cluster center in turn ba , and the calculation formula is expressed as: where D ba is the distance from the b-th calculation point j b to the a-th cluster center z a , where b = 1, 2, 3, …, m - 3; a = 1, 2, 3; R is the dimension of the sample point, u = 1, 2, 3, …, R; j bu is the coordinate of the b-th calculation point j b in the u-th dimension, and z au is the coordinate of the a-th cluster center z a in the u-th dimension; m is the total number of data, and each data corresponds to a sample point; Step S13: Establish three corresponding clusters based on the three clustering centers; Step S14: Compare the distances from the calculation point j b to each cluster center, and assign the calculation point to the cluster corresponding to the cluster center with the closest distance; Step S15: Let b = b + 1, and jump back to Step S14; Step S16: Repeat Step S14 to Step S15 until the loop ends when b = m - 3, and allocate the m - 3 calculation points to the corresponding clusters; Step S17: Recalculate the coordinates of the new cluster center points for each cluster. The calculation formula is expressed as: where z′ c is the coordinate of the new cluster center point of the c-th cluster, c = 1, 2, 3, y i is the coordinate of the i-th calculation point in the c-th cluster, i = 1, 2, 3, …, I, and I is the total number of calculation points in the c-th cluster, y i =(y i1 , y i2 , y i3 ,..., y iR ); Step S18: Repeat Step S12 to Step S17 until the new clustering center point coordinates of each cluster are recalculated in Step S17, and the loop ends when they are the same as the new clustering center point coordinates of each cluster calculated in the previous loop; Obtain the clustering corresponding to the three clusters and the corresponding sample points, and use the data type of the clustering center of each cluster as the data cluster type of the cluster. The sample points in the cluster are all data type data features corresponding to the cluster.

4. The energy-saving optimization method for base stations in a power wireless communication network according to claim 3, characterized in that: The specific method for analyzing the data types of outliers in Step S03 is: Annotate the real-time energy consumption data of n energy devices of the base station, and annotate the energy consumption of the f-th energy device at the t-th moment as θ ft , where f = 1, 2, 3, …, n, t = 1, 2, 3, …, T, and T is the total number of moments; Obtain the average energy consumption of n energy devices, and the calculation formula is expressed as: Among them, P f is the average energy consumption of the f-th energy device; Values higher than the average energy consumption are calibrated as outliers, and the outliers of each energy device form an outlier set U. The outlier set of the f-th energy device is expressed as: U f ={uf1, uf2, uf3,..., uf q}, where U f is the outlier set of the f-th energy device, and uf q is the q-th outlier data in the f-th energy device; Obtain the outlier sets of n energy devices; Calculate the distances from the data characteristics of each outlier in the outlier set of each energy device to the new clustering center points of each cluster in step S18, and arrange them in ascending order from small to large. Assign each outlier data to the cluster corresponding to the closest clustering center point, that is, the cluster corresponding to the clustering center point with the smallest distance value. The data type corresponding to this cluster is the data type of the outlier data assigned to this cluster; The calculation formula is as follows: where Df qc is the distance between the q-th outlier data in the outlier set of the f-th energy device and the new clustering center point of the c-th cluster, fq u is the coordinate of the q-th outlier data in the outlier set of the f-th energy device in the u-th dimension, z′ cu is the new clustering center point z′ of the c-th cluster c in the coordinate of the u-th dimension, R is the dimension in step S12, and u = 1, 2, 3, …, R; Obtain the data types of each outlier in the outlier set of each energy device.

5. The base station energy-saving optimization method in a power wireless communication network according to claim 4, wherein: The specific method of mapping the data type of the outlier to the energy device in step S04 is as follows: Establish a one-to-one mapping relationship between each outlier set of energy devices and the corresponding energy device. Calculate the proportion of each outlier data type in each outlier set of energy devices. Add the proportion of the fault feature data type and the proportion of the aging feature data type. If it exceeds the set threshold, record the energy device mapped by this outlier set as the target energy device; The formula is expressed as: Among them, Zf1 is the proportion of the failure feature data type in the f-th outlier set of energy equipment, Zf2 is the proportion of the aging feature data type in the f-th outlier set of energy equipment, Zf3 is the proportion of the normal fluctuation feature data type in the f-th outlier set of energy equipment, and Qf is the number of all outliers in the f-th outlier set of energy equipment; If Zf1 + Zf2 ≥ YU, record the energy device mapped by this outlier set as the target energy device, where YU is the set threshold.

6. The base station energy-saving optimization method in a power wireless communication network according to claim 5, characterized in that: The method for establishing the grid model is as follows: According to the physical structure and components of each target energy device, discretize the target energy device into a grid structure. The grid structure contains several grid cells, and each grid cell represents a physical structure or component of the target energy device; Set a discrete state for each grid cell, such as normal state, aging state, and fault state; The discrete state is represented by binary coding. The state of the grid cell can be defined as {0: normal state; 1: aging state; 2: fault state}; Define the state transition rule of the w-th grid cell and represent the state transition rule with a Boolean function; This grid structure is the grid model; The state transition rule is: If itself or any adjacent grid cell is in the fault state, then transfer to the fault state; Otherwise, if itself or any adjacent grid cell is in the aging state, then transfer to the aging state.

7. A base station energy-saving optimization method in a power wireless communication network according to claim 6, characterized in that: The specific method for predicting the outlier rate of the target energy device based on the constructed grid model is as follows: For each target energy device, update the state of each of its grid cells until the current moment to obtain the real-time grid model; Based on the overall grid state of the real-time grid model, update the overall grid state at the next moment, obtain the failure rate and aging rate in the overall grid state at the next moment, and get the outlier rate of the target energy device at the next moment; The failure rate is the proportion of grid cells in the fault state in the overall grid state at the next moment; The aging rate is the proportion of grid cells in the aging state in the overall grid state at the next moment; Add the value obtained by multiplying the failure rate by the corresponding weight factor and the value obtained by multiplying the aging rate by the corresponding weight factor to get the outlier rate; Thus, obtain the outlier rate of each target energy device.

8. A base station energy-saving optimization method in a power wireless communication network according to claim 7, characterized in that: The method of state update is as follows: For the e-th time step, traverse all grid cells. For the w-th grid cell, obtain the states of itself and its adjacent grid cells at the (e - 1)-th time step; calculate the state of the w-th grid cell at the e-th time step according to the state transition rule, and update the state of the w-th grid cell. Repeat and record the states of all grid cells at the e-th time step, which is the overall grid state, to form a grid model; the time step is the time difference between the current moment and the previous moment or the next moment.

9. The base station energy-saving optimization method in a power wireless communication network according to claim 8, characterized in that: The best energy-saving optimization measures are obtained by using an improved nature-inspired algorithm, specifically: Step S21: Encode the energy-saving optimization set, and the encoding is O. O is the chromosome. Obtain the chromosome and construct the initial population A = {O1, O2, O3,..., O k}; Step S22: Determine the fitness function; Step S23: Conduct natural selection on the chromosomes in the population; Step S24: Conduct crossover recombination on the chromosomes in the population; Step S25: Conduct mutation on the chromosomes in the population; Step S26: Obtain a new population. Preset the population generation number as L and the fitness threshold as Q, where L is an integer greater than 0 and Q is a real number greater than 0. Loop from Step S23 to Step S25 until the generation number of the new population is L or there exists a chromosome in the new population whose corresponding fitness is greater than or equal to the fitness threshold Q. Then the loop ends. Obtain the energy-saving optimization set corresponding to the chromosome with the maximum fitness in the new population as the best energy-saving optimization set, and the energy-saving optimization measures in the best energy-saving optimization set are the best energy-saving optimization measures.

10. A method for optimizing the energy saving of a base station in a power wireless communication network according to claim 9, characterized in that: The expression of the fitness function is: λ r =-YZ r , where λ r is the fitness corresponding to the r-th chromosome, and YZ r is the abnormal rate suppression value of the energy-saving optimization set corresponding to the r-th chromosome; r = 1, 2, 3, …, k; The way to obtain the anomaly rate suppression value is: Take the energy-saving optimization measures in the energy-saving optimization set as input data, input them into the energy-saving optimization measure prediction model, obtain the anomaly rates of each energy device corresponding to the energy-saving optimization measures in the energy-saving optimization set as output, perform weighted averaging on the output energy device anomaly rates to obtain the average anomaly rate, and subtract the predicted average anomaly rate of the energy device to obtain the anomaly rate suppression value; The construction method of the energy-saving optimization measure prediction model is: Take the energy-saving optimization measures in the energy-saving optimization set, the energy consumption data of the corresponding energy devices after energy-saving optimization, and the device operation parameter data as analysis data. Pre-collect d sets of analysis data, where d is an integer greater than 1. Convert the analysis data and the corresponding energy device anomaly rates into a corresponding set of feature vectors; each set of analysis data is the energy-saving optimization measures of an energy device and the corresponding energy consumption data of the energy device after energy-saving optimization and the device operation parameter data. Use each set of feature vectors as the input of the energy-saving optimization measure prediction model. The energy-saving optimization measure prediction model outputs a set of predicted energy equipment abnormality rates corresponding to each set of analysis data, uses the actual energy equipment abnormality rate corresponding to each set of analysis data as the prediction target, and the actual energy equipment abnormality rate is the energy equipment abnormality rate collected in advance corresponding to the analysis data; uses minimizing the sum of prediction errors of all analysis data as the training target; the formula for the prediction error is expressed as: ε p =σ p -μ p , where ε p is the prediction error, p is the group number of the feature vector corresponding to the analysis data, σ p is the predicted energy equipment abnormality rate corresponding to the p-th group of analysis data, μ p is the actual energy equipment abnormality rate corresponding to the p-th group of analysis data. Train the energy-saving optimization measure prediction model until the sum of prediction errors reaches convergence and then stop training.

Images