Coal mine data processing method and system based on cloud edge collaboration

Abstract

The invention relates to the technical field of intelligent mine Internet of Things and edge intelligent computing, in particular to a coal mine data processing method and system based on cloud-edge collaboration, and the method comprises the steps: collecting coal mine data, and packaging the data into an entropy energy data capsule; constructing a virtual osmotic membrane model, and calculating an osmotic pressure difference as a priority by combining channel flow resistance and model difference; if the fingerprint is larger than a threshold value, the fingerprint is uploaded to the cloud, and otherwise, local crystallization storage is executed and the fingerprint is synchronized; the cloud updates the model and feeds back the parameter difference, and the edge dynamically adjusts the pressure difference calculation weight according to the parameter difference; according to the method, the security retention and cloud indexability of intercepted data are ensured, automatic balance between calculation task unloading and network transmission protection is realized, and the robustness of the system is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart mine Internet of Things and edge intelligent computing, in particular to a coal mine data processing method and system based on cloud-edge collaboration. BACKGROUND

[0002] In the complex industrial Internet of Things environment of the coal mine underground, a large number of sensors continuously generate high-frequency time series data, which usually contains multi-dimensional information such as vibration waveform and environmental parameters, and the network transmission environment has characteristics such as limited bandwidth, severe signal fluctuation and variable channel flow resistance;

[0003] In order to realize the monitoring of the underground state, the existing data processing scheme usually adopts a simple threshold trigger upload or full data transparent transmission mode; the full upload mode is prone to network congestion and data backlog in a weak network environment, causing high transmission delay and packet loss; while the filtering mode based on fixed physical threshold can reduce the amount of transmission data, but it is difficult to identify potential risk characteristics with high information entropy or large cognitive difference with the cloud model when the value does not exceed the limit, resulting in the loss of key zero-sample hidden danger data; in addition, the traditional scheduling strategy often separates the real-time load of the edge computing node and the cognitive iteration demand of the cloud global model, lacking overall analysis of data value density, network real-time flow resistance and edge computing power potential; such static and single-dimensional scheduling method cannot effectively adapt to the limited network bandwidth, limited edge computing power and massive data upload demand, making it difficult to balance the real-time, integrity of data transmission and load balancing of edge nodes when facing sudden high-frequency data flow, and unable to effectively solve the resource deadlock problem between high-value data screening and inefficient transmission in a weak network environment; therefore, how to build a dynamic scheduling mechanism that can adaptively perceive data information entropy, model cognitive difference and network flow resistance to realize efficient data processing and accurate shunting under cloud-edge collaboration has become a technical problem to be solved. SUMMARY

[0004] The purpose of the present application is to provide a coal mine data processing method and system based on cloud-edge collaboration, specifically, the technical solution of the present application is as follows:

[0005] A coal mine data processing method based on cloud-edge collaboration, running in a heterogeneous system including edge computing nodes and cloud servers, the method comprising:

[0006] Collecting coal mine underground sensor data and encapsulating the sensor data into a data packet containing local information entropy parameters at the edge computing node, defining the data packet as an entropy energy data capsule;

[0007] A scheduling model based on virtual penetration membrane logic is constructed, the flow resistance parameters of the current network channel and the model parameter cognitive difference degree between the edge computing node and the cloud server are obtained, the transmission priority value of the entropy-energy data capsule relative to the cloud server is calculated, and the transmission priority value is defined as the osmotic pressure difference;

[0008] A data shunting strategy based on the osmotic pressure difference is executed: the size relationship between the osmotic pressure difference and a preset threshold is determined; if the osmotic pressure difference is greater than the preset threshold, the entropy-energy data capsule is sent to the cloud server through the network; if the osmotic pressure difference is less than or equal to the preset threshold, local compression storage is performed on the entropy-energy data capsule and an index is generated, the process is defined as local crystallization storage, and the data fingerprint in the index is synchronized to the cloud server;

[0009] The cloud server updates the global model according to the received entropy-energy data capsule, and feeds back the parameter difference between the global model and the local model to the edge computing node, and the edge computing node dynamically adjusts the weight coefficient when calculating the osmotic pressure difference subsequently according to the parameter difference.

[0010] Optionally, the data structure of the entropy-energy data capsule includes original waveform data, local information entropy, thermal attenuation coefficient and edge computing power potential energy.

[0011] The method for packaging the entropy-energy data capsule containing the local information entropy parameter includes:

[0012] Obtaining the original waveform data, calculating the data distribution sparsity value of the original waveform data, and marking the data distribution sparsity value as local information entropy;

[0013] Obtaining the difference between the timestamp of data acquisition and the current system time, calculating the attenuation factor changing with time based on the difference, and marking the attenuation factor as the thermal attenuation coefficient;

[0014] Obtaining the current CPU idle rate of the edge computing node, calculating the reciprocal of the CPU idle rate, and marking the reciprocal as the edge computing power potential energy;

[0015] Combining the original waveform data, local information entropy, thermal attenuation coefficient and edge computing power potential energy to generate the entropy-energy data capsule.

[0016] Optionally, the method for calculating the osmotic pressure difference of the entropy-energy data capsule relative to the cloud server includes:

[0017] Obtaining the global model parameters fed back by the cloud server and the current model parameters of the edge computing node, calculating the gradient residual between the parameter vectors of the two, and marking the gradient residual as the cognitive difference degree;

[0018] Obtaining a bandwidth occupancy rate and a packet loss rate of a current network channel, and calculating a network flow resistance parameter based on the bandwidth occupancy rate and the packet loss rate;

[0019] Obtaining a first weighting coefficient, a second weighting coefficient and a third weighting coefficient;

[0020] Multiplying the local information entropy by the first weighting coefficient to obtain an entropy energy term;

[0021] Multiplying the cognitive difference by the second weighting coefficient to obtain a potential energy term;

[0022] Multiplying the network flow resistance parameter by the third weighting coefficient to obtain a resistance term;

[0023] Calculating the sum of the entropy energy term and the potential energy term, and then subtracting the resistance term, to obtain a calculation result as the osmotic pressure difference.

[0024] Optionally, the method for performing local crystallization storage on the entropy energy data capsule comprises:

[0025] Compressing the original waveform data in the entropy energy data capsule by using a lossless compression algorithm to generate a compressed data block;

[0026] Writing the compressed data block into a non-volatile memory of an edge computing node;

[0027] Performing a hash operation on the entropy energy data capsule to generate a unique identification code, and marking the unique identification code as a data fingerprint;

[0028] Constructing an index data packet containing the data fingerprint and the osmotic pressure difference, and sending the index data packet to a cloud server.

[0029] Optionally, the method further comprises:

[0030] The cloud server receives index data packets from multiple edge computing nodes;

[0031] Parsing the index data packet to obtain the osmotic pressure difference corresponding to the data fingerprint and sensor geographic location information;

[0032] Mapping the sensor geographic location information to a mine space coordinate system, and constructing a full-mine information pressure field distribution map by taking the osmotic pressure difference as a thermal value;

[0033] Monitoring a high-pressure area in the full-mine information pressure field distribution map, and if a sensor in the high-pressure area does not trigger a conventional threshold alarm but the average osmotic pressure difference in the area continues to rise, a zero-sample hidden danger early warning signal is generated.

[0034] Optionally, the method for updating the global model comprises:

[0035] The cloud server unpacks the received entropy data capsule and parses out original waveform data;

[0036] The original waveform data is used to incrementally train a current global model to obtain updated global model parameters;

[0037] The change amount of the updated global model parameters and the last version model parameters is calculated;

[0038] If the change amount is greater than a preset convergence threshold, a model update instruction is generated, and the updated global model parameters are issued to the edge computing node.

[0039] Optionally, the method for dynamically adjusting the subsequent osmotic pressure difference calculation weight comprises:

[0040] The edge computing node receives the model update instruction and parses out the updated global model parameters;

[0041] The updated global model parameters are used to replace the local model parameters, and the parameter update frequency is recorded;

[0042] The second weighting coefficient is adjusted based on the parameter update frequency: if the parameter update frequency increases, the value of the second weighting coefficient is increased to increase the weight of the cognitive difference in the osmotic pressure difference calculation; if the parameter update frequency decreases, the value of the second weighting coefficient is decreased.

[0043] Optionally, collecting coal mine underground sensor data comprises:

[0044] Through the vibration sensors and gas sensors deployed in the coal mining face and the roadway, vibration waveform data and gas concentration time series data are collected in real time;

[0045] The vibration waveform data and the gas concentration time series data are denoised and preprocessed to eliminate invalid data segments caused by equipment electromagnetic interference.

[0046] A coal mine data processing system based on cloud-edge collaboration is used to implement the coal mine data processing method based on cloud-edge collaboration as claimed in any one of claims 1-8, comprising:

[0047] A data acquisition and encapsulation module configured to acquire coal mine underground sensor data and encapsulate the sensor data into a data packet containing local information entropy parameters as an entropy data capsule at an edge computing node;

[0048] An osmotic model construction module configured to construct a scheduling model based on a virtual osmotic membrane logic, calculate the transmission priority value of the entropy data capsule relative to the cloud server as an osmotic pressure difference based on the flow resistance parameters of the current network channel and the model parameter cognitive difference between the edge computing node and the cloud server;

[0049] The shunting execution module is configured to execute a data shunting strategy based on the osmotic pressure difference: if the osmotic pressure difference is greater than a preset threshold, the entropy energy data capsule is sent to a cloud server through a network; if the osmotic pressure difference is less than or equal to the preset threshold, local compression storage is performed on the entropy energy data capsule, and an index is generated, the process is defined as local crystallization storage, and data fingerprints are extracted and synchronized to the cloud server;

[0050] The cloud feedback module is configured to update a global model according to the received entropy energy data capsule, and feed back model parameter differences to the edge computing node, so that the edge computing node dynamically adjusts weight coefficients when calculating the osmotic pressure difference subsequently.

[0051] Compared with the prior art, the present application has the following beneficial effects:

[0052] 1. The present application converts the complex network service quality scheduling into a convex optimization problem in physics by constructing the entropy energy data capsule and the virtual osmotic membrane scheduling model; the method comprehensively considers the data local information entropy, model cognitive difference and network channel flow resistance, and accurately calculates the transmission priority; in the weak network environment with limited bandwidth and severe fluctuations in the coal mine underground, the system can adaptively select the data with the most information value for priority uploading, which not only ensures that the key disaster precursor characteristics are not lost, but also avoids invalid data blocking the channel, effectively solving the resource deadlock problem between massive high-frequency data uploading and limited bandwidth;

[0053] 2. The present application introduces model parameter cognitive difference and local information entropy as evaluation indexes, breaking through the limitation of traditional fixed physical threshold filtering data; the system can identify potential hidden danger characteristics with high heterogeneity or high uncertainty although the numerical value does not exceed the limit, realizing sensitive capture of zero-sample risk mode; this not only ensures that the cloud global model can obtain high-value samples for iterative update in time, but also significantly improves the perception sensitivity and foresight of the system to unknown disaster precursors in complex working conditions;

[0054] 3. The present application establishes an adaptive load balancing and local crystallization storage mechanism based on edge computing power potential; the system dynamically adjusts the shunting threshold according to the CPU idle rate of the edge node, and automatically performs lossless compression storage on the data and only synchronizes the lightweight fingerprint index when the computing load is too high or the network is congested; this mechanism not only prevents edge hardware overload, but also ensures the safe retention of intercepted data and the indexability of the cloud, realizes the automatic balance between computing task offloading and network transmission protection, and greatly improves the robustness of the system;

[0055] 4. The present application uses the uploaded lightweight index data packet to construct the whole mine information pressure field, realizes the macroscopic situation awareness independent of network quality; the cloud server maps the sensor geographic position and the osmotic pressure difference thermodynamic value, and realizes real-time monitoring of the spatial distribution and accumulation trend of the underground information pressure; even if the original waveform data is retained locally due to network reasons, the system can still identify the atypical risk mode in advance by analyzing the high-pressure area change rate of the pressure field, realize the leap from the traditional point threshold alarm to the planar trend warning. BRIEF DESCRIPTION OF DRAWINGS

[0056] The present application will be further explained in conjunction with the accompanying drawings and embodiments:

[0057] Figure 1 is a flow chart of the method of the present application;

[0058] Figure 2 is a structural diagram of the system of the present application. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail in conjunction with specific embodiments.

[0060] Example 1:

[0061] Please refer to Figure 1 A coal mine data processing method based on cloud edge collaboration, running in a heterogeneous system including edge computing nodes and cloud servers, the method comprising:

[0062] Collecting coal mine underground sensor data, and encapsulating the sensor data into a data packet containing local information entropy parameters in the edge computing node, defining the data packet as an entropy energy data capsule;

[0063] Building a scheduling model based on virtual penetration membrane logic, obtaining the flow resistance parameters of the current network channel and the model parameter cognitive difference degree between the edge computing node and the cloud server, calculating the transmission priority value of the entropy energy data capsule relative to the cloud server, defining the transmission priority value as the osmotic pressure difference;

[0064] Performing data shunting strategy based on osmotic pressure difference: determining the size relationship between the osmotic pressure difference and the preset threshold; if the osmotic pressure difference is greater than the preset threshold, the entropy energy data capsule is sent to the cloud server through the network; if the osmotic pressure difference is less than or equal to the preset threshold, the entropy energy data capsule is locally compressed and stored and an index is generated, the process is defined as local crystallization storage, and the data fingerprint in the index is extracted and synchronized to the cloud server;

[0065] The cloud server updates the global model according to the received entropy-energy capsule, and feeds back the parameter difference between the global model and the local model to the edge computing node, and the edge computing node dynamically adjusts the weight coefficient when calculating the osmotic pressure difference in the subsequent calculation according to the parameter difference.

[0066] The embodiment provides a solution to the deadlock problem between mass high-frequency sensor data uploading and limited bandwidth in a weak network environment in a coal mine underground;

[0067] The system performs the step of collecting sensor data in a coal mine, and obtains sensor stream data in real time through an industrial bus;

[0068] Inside the edge computing node, the sensor data is encapsulated into a data packet containing local information entropy parameters, and the data packet is defined as an entropy-energy capsule;

[0069] Here, the entropy-energy capsule refers to a standardized data container carrying thermodynamic properties, which not only contains original physical observation values, but also carries metadata describing the value density and transmission urgency of the data;

[0070] The system constructs a scheduling model based on a virtual osmotic membrane logic, which is deployed in the memory of the edge node, and obtains the flow resistance parameters of the current network channel and the model parameter cognitive difference between the edge computing node and the cloud server in real time;

[0071] Flow resistance parameters: derived from network environment monitoring, and physically representing the degree of hindrance of the current network environment to data transmission;

[0072] Model parameter cognitive difference: derived from model comparison calculation, and physically representing the cognitive bias between the AI model currently held by the edge and the latest global model in the cloud;

[0073] Based on the above parameters, the system calculates the transmission priority value of the entropy-energy capsule relative to the cloud server, and defines the transmission priority value as the osmotic pressure difference;

[0074] The system executes a data shunting strategy based on the osmotic pressure difference: determines the size relationship between the osmotic pressure difference and the preset threshold;

[0075] In response to the osmotic pressure difference being greater than the preset threshold, the system determines that the entropy-energy capsule has a very high information entropy or a large current cognitive difference, and sends the entropy-energy capsule to the cloud server through the network;

[0076] In response to the osmotic pressure difference being less than or equal to the preset threshold, the system determines that the data is mediocre or the network is congested, performs local compression storage on the entropy-energy capsule, and generates an index, defines the process as local crystallization storage, and synchronizes the data fingerprint in the index to the cloud server.

[0077] The cloud server updates the global model according to the received entropy energy data capsule, and feeds back the parameter difference between the global model and the local model to the edge computing node, and the edge computing node dynamically adjusts the weight coefficient when calculating the osmotic pressure difference in the subsequent calculation according to the parameter difference;

[0078] The embodiment introduces the concepts of entropy energy data capsule and osmotic pressure difference, and converts the complex QoS scheduling problem into a convex optimization problem in physics; in the coal mine underground environment with limited and volatile network bandwidth, this method can adaptively select the data with the most information value for uploading, which not only ensures that the key disaster precursor characteristics are not lost, but also avoids invalid data blocking the network channel, and maximizes the cloud-edge collaborative efficiency.

[0079] Embodiment 2:

[0080] The data structure of the entropy energy data capsule includes original waveform data, local information entropy, thermal decay coefficient and edge computing power potential energy;

[0081] The method of encapsulating the entropy energy data capsule containing the local information entropy parameter includes: obtaining the original waveform data, calculating the data distribution sparsity value of the original waveform data, and marking the data distribution sparsity value as the local information entropy;

[0082] Obtain the difference between the timestamp of data acquisition and the current system time, calculate the decay factor changing with time based on the difference, and mark the decay factor as the thermal decay coefficient;

[0083] Obtain the current CPU idle rate of the edge computing node, calculate the reciprocal of the CPU idle rate, and mark the reciprocal as the edge computing power potential energy;

[0084] Combine the original waveform data, the local information entropy, the thermal decay coefficient and the edge computing power potential energy to generate the entropy energy data capsule.

[0085] The embodiment is a further embodiment of the entropy energy data capsule structure in embodiment 1;

[0086] The system obtains the original waveform data, calculates the data distribution sparsity value of the original waveform data, and marks the data distribution sparsity value as the local information entropy; in order to solve the logical inverse relationship between the more sparse the zero value is, the more important the data value is, and the more complex the data value is, the more important the data value is;

[0087] The data distribution sparsity value defined in the embodiment Specific information density or non-sparsity, the calculation formula is Or use It should be noted that the local information entropy defined in the application specifically refers to a signal complexity index based on sparsity representation, and its physical meaning is to measure the non-stationary degree and mutation characteristics of the waveform, which is different from the calculation method of the traditional Shannon information entropy in the sense of probability statistics, but both are used to represent the value density of information.

[0088] Among them, represents a discrete vector composed of original waveform data;

[0089] : represents the total number of data points in the sampling window;

[0090] , , : respectively represent the L0, L1 and L2 norms of the vector;

[0091] It should be particularly noted that when using the formula , in order to avoid the constant equal to , that is, almost no absolute zero value, an effective signal threshold is introduced, for example, the value is 1.5 times the amplitude of the environmental background noise, only the data points with an absolute value greater than are counted as non-zero elements, so that the more complex and more mutated the waveform is, that is, the lower the physical sparsity is, the higher the information density is, the greater the value is, thereby generating greater entropy force in subsequent calculation;

[0092] The system obtains the difference between the data acquisition timestamp and the current system time, calculates the decay factor changing with time based on the difference, and marks the decay factor as a thermal decay coefficient;

[0093] The specific calculation formula of the coefficient is:

[0094]

[0095] Among them, : from the system clock, the physical meaning is the current system time;

[0096] : from the data packet header, the physical meaning is the data acquisition timestamp;

[0097] : from the preset configuration, the physical meaning is the time constant, for example, the value is 500ms;

[0098] The system obtains the current CPU idle rate of the edge computing node, calculates the reciprocal of the CPU idle rate, and marks the reciprocal as the edge computing power potential energy; the edge computing power potential energy The calculation formula is:

[0099]

[0100] The edge computing node releases data to the outside due to the accumulation of computing load, and the edge computing node has the ability to release data to the outside. The higher the load is, the greater the potential energy is, and the stronger the driving force of data overflow is.

[0101] wherein, : the source is a system monitoring interface, and the physical meaning is the current CPU idle rate between 0 and 1;

[0102] : the source is a constant definition, for example , the physical meaning is a small amount to prevent the denominator from being zero;

[0103] The system combines the original waveform data, local information entropy, thermal decay coefficient and edge computing power potential energy to generate an entropy energy data capsule.

[0104] The embodiment unifies the time value of data, thermal decay, content value, information entropy, i.e. the corrected information density and environmental constraints, and computing power potential energy, so that data scheduling no longer depends only on data size or source, but is based on the multi-dimensional comprehensive value of data, significantly improving the fine degree of scheduling, and ensuring that when the edge node is under high load or data failure, a discard or upload decision that conforms to the physical law can be made.

[0105] Embodiment 3:

[0106] The method for calculating the osmotic pressure difference of the entropy energy data capsule relative to the cloud server includes:

[0107] Obtain the global model parameters fed back by the cloud server and the current model parameters of the edge computing node, calculate the gradient residual between the two parameter vectors, and mark the gradient residual as the cognitive difference degree;

[0108] Obtain the bandwidth occupancy rate and packet loss rate of the current network channel, and calculate the network flow resistance parameter based on the bandwidth occupancy rate and the packet loss rate;

[0109] Obtain the first weighting coefficient, the second weighting coefficient and the third weighting coefficient;

[0110] Multiply the local information entropy by the first weighting coefficient to obtain the entropy energy term;

[0111] Multiply the cognitive difference degree by the second weighting coefficient to obtain the gradient potential energy term, and multiply the network flow resistance parameter by the third weighting coefficient to obtain the resistance term;

[0112] Calculate the sum of the entropy energy term and the gradient potential energy term, and then subtract the resistance term, and take the obtained calculation result as the osmotic pressure difference.

[0113] This embodiment is a further specification of the osmotic pressure difference calculation logic in Embodiments 1 and 2;

[0114] The system obtains the global model parameters fed back by the cloud server and the current model parameters of the edge computing node, calculates the gradient residual between the two parameter vectors, and marks the gradient residual as the cognitive difference.

[0115] Specifically, the model parameters are expanded into a one-dimensional vector, and the cognitive dissimilarity is calculated using Euclidean distance, i.e., the L2 norm.

[0116]

[0117] in This represents the parameter vector; to ensure that this difference is comparable in scale to the local information entropy 0-1 interval, and to avoid problems caused by the large number of model parameters... If the numerical value is too large and dominates the calculation result, this step must be performed... Normalization is performed to ensure that it is consistent with the local information entropy. They are comparable in the 0-1 dimension interval, specifically mapped using the Sigmoid function, with the calculation formula as follows:

[0118]

[0119] in, The scaling factor is used to control the slope of the curve. This is the center offset; to ensure the mapping function remains sensitive to changes in model differences and avoids entering the saturation region, Value is historical A time window is selected in this embodiment. That is, within the most recent N periods The moving average is given by the formula: ; The dimensions are The reciprocal of the dimensions, its value is set as ,in Within the historical cycle standard deviation For example, a preset minimum constant. This is used to prevent division-by-zero overflow errors when the standard deviation approaches zero, thus achieving standardized data input; after this mapping, The range of this scalar value is strictly constrained to the interval (0,1); It accurately reflects the geometric distance between the edge model and the global optimal solution in the parameter space;

[0120] The system obtains the current network channel bandwidth occupancy and packet loss rate, and calculates network flow resistance parameters based on the bandwidth occupancy and packet loss rate;

[0121] The network flow resistance parameter calculation formula is:

[0122]

[0123] wherein, and are preset dimensionless normalization weight coefficients, for example, take , and satisfy ;

[0124] : is the normalized bandwidth occupancy rate, the calculation method is the current instantaneous network throughput divided by the theoretical maximum bandwidth capacity of the network channel, the value range is 0-1;

[0125] : is the normalized packet loss rate, the calculation method is the number of lost packets in the current sampling period divided by the total number of sent data packets, the value range is 0-1;

[0126] The system obtains a first weighting coefficient , a second weighting coefficient and a third weighting coefficient , and performs weighted calculation: the local information entropy is multiplied by to obtain the entropy item; the cognitive difference degree is multiplied by to obtain the potential energy item; the network flow resistance parameter is multiplied by to obtain the resistance item;

[0127] Here, it is particularly pointed out that the potential energy item calculated in this step is transformed from the cognitive difference degree, and specifically refers to the model gradient potential energy, in order to distinguish from the edge computing potential energy based on hardware resources defined in Embodiment 2; although both borrow the concept of potential energy, the former describes the potential driven by knowledge difference, and the latter describes the absorbing potential generated by idle computing resources;

[0128] At the same time, the thermal attenuation coefficient defined in Embodiment 2 is used here to modify the first weighting coefficient in real time, that is, , wherein is the initial preset value of the first weighting coefficient, is the real-time entropy weight after introducing time attenuation; so as to ensure that the entropy driving force of data naturally attenuates over time, reflecting the timeliness of data;

[0129] The sum of the entropy item and the gradient potential energy item is calculated, and then the resistance item is subtracted, and the obtained calculation result is taken as the osmotic pressure difference; the specific calculation model is:

[0130]

[0131] The physical meaning of the formula is that the positive entropy energy and the gradient potential energy jointly constitute the driving force for uploading, and the network flow resistance constitutes the reverse resistance. Only when the driving force is greater than the resistance, the data has the qualification to penetrate the uploading;

[0132] In order to realize adaptive shunting based on edge computing power load, the preset threshold for determining the size relationship in this step is not a fixed value, but is based on the edge computing power potential energy calculated in embodiment 2 The dynamic threshold is dynamically adjusted The calculation formula is:

[0133]

[0134] Among them, : the reference threshold, derived from the initial configuration of the system, for example, the value is 0.5;

[0135] : the load sensitivity coefficient, for example, the value is 0.3;

[0136] : the normalized upper limit value of the edge computing power potential energy; based on the definition of the edge computing power potential energy in embodiment 2 , the value is , that is, the theoretical limit potential energy value when the CPU idle rate tends to 0; The physical meaning of the formula is that when the edge node CPU load increases, the idle rate decreases,

[0137] the value increases, so that the dynamic threshold decreases; this will make more data meet the condition of and be shunted to the cloud, thereby reducing the computing and storage pressure of the edge node, and realizing the load balancing of cloud-edge collaboration; the premise of this logic is that the resistance term calculated based on embodiment 3 does not reach the blocking threshold, that is, the system preferentially relieves the edge pressure by unloading computing tasks, but if the network congestion is serious, the resistance term is too large, and the physical formula will naturally stop uploading, so as to achieve automatic balance between computing unloading and network protection;

[0138] ​The embodiment establishes a dynamic balance mechanism, converts the abstract model difference and network state into a calculable scalar by introducing L2 norm and normalized weighting, and generates a positive thrust to overcome even great network resistance and force data upload when data is extremely abnormal, i.e., high entropy energy or model is extremely uncertain, i.e., high gradient potential energy; on the contrary, if the network resistance is too large, only data with great value can pass through, and ordinary data is left in the local, thereby realizing optimal data throughput control in a dynamic network environment.

[0139] Embodiment 4:

[0140] The method for performing local crystallization storage on the entropy energy data capsule comprises:

[0141] The original waveform data in the entropy energy data capsule is compressed by using a lossless compression algorithm to generate a compressed data block;

[0142] The compressed data block is written into the non-volatile memory of the edge computing node;

[0143] Hash operation is performed on the entropy energy data capsule to generate a unique identification code, and the unique identification code is marked as a data fingerprint;

[0144] An index data packet containing the data fingerprint and the osmotic pressure difference is constructed, and the index data packet is sent to the cloud server.

[0145] The embodiment is a further embodiment of the local storage strategy performed when the data does not pass through the penetration membrane in embodiment 3;

[0146] The system compresses the original waveform data in the entropy energy data capsule by using a lossless compression algorithm to generate a compressed data block;

[0147] In this process, the LZ4 or Zstandard algorithm is used for high-speed lossless compression, and the originally flowing data is frozen into a high-density storage block;

[0148] The system writes the compressed data block into the non-volatile memory of the edge computing node, and the storage medium is selected as an industrial-grade NVMeSSD to ensure the safety of data in power failure;

[0149] The system performs hash operation on the entropy energy data capsule to generate a unique identification code, and the unique identification code is marked as a data fingerprint; the original data is calculated by using the SHA-256 algorithm;

[0150] Calculate the intrinsic information pressure , and the calculation formula is:

[0151]

[0152] The numerical value eliminates the influence of network flow resistance and only represents the physical value of data itself and the difference between the model;

[0153] The index data packet is sent to the cloud server, and the index data packet is sent to the cloud server;

[0154] The crystallization storage mechanism ensures that even if the data is not uploaded, the cloud server can know the existence of the data through the received data fingerprint, and the importance of the data through the osmotic pressure difference value in the index; This not only greatly saves the uplink bandwidth, and the compression ratio can reach more than 10:1, but also provides a solid index foundation for subsequent cloud on-demand backtracking, such as accident backtracking according to the fingerprint to retrieve local data.

[0155] Embodiment 5:

[0156] The method further comprises:

[0157] The cloud server receives the index data packet from the plurality of edge computing nodes;

[0158] The index data packet is parsed to obtain the osmotic pressure difference corresponding to the data fingerprint and the sensor geographic location information;

[0159] The sensor geographic location information is mapped into a mine space coordinate system, and the intrinsic information pressure is taken as a thermal value to construct a mine-wide information pressure field distribution map; This is done to ensure that the risk assessment is independent of network transmission quality;

[0160] The high-pressure area in the mine-wide information pressure field distribution map is monitored, and if the sensor in the high-pressure area does not trigger a regular threshold alarm, but the average osmotic pressure difference in the area continues to rise, a zero-sample hidden danger early warning signal is generated.

[0161] This embodiment describes in detail the mine-wide information pressure field and early warning method based on the index data packet;

[0162] The cloud server receives the index data packet from the plurality of edge computing nodes, and parses the index data packet to obtain the osmotic pressure difference corresponding to the data fingerprint and the sensor geographic location information;

[0163] The system maps the sensor geographic location information into a mine space coordinate system, and takes the osmotic pressure difference as a thermal value to construct a mine-wide information pressure field distribution map; In the mine digital twin map maintained in the cloud, the osmotic pressure difference value uploaded by each sensor is mapped as the color depth of the location;

[0164] The system monitors the high-pressure area in the mine-wide information pressure field distribution map;

[0165] The specific monitoring method is as follows:

[0166] Space gridding: dividing the mine space into grid cells, where represents the number of discrete intervals divided along the tunnel direction, represents the number of discrete intervals divided along the vertical section of the tunnel, thus forming a two-dimensional space grid; wherein is a grid dimension parameter that distinguishes from the number of sampling points in Example 2, the arithmetic mean of the osmotic pressure difference of all sensors in each grid cell is calculated, denoted as ;

[0167] Region determination: if the of a grid cell is greater than the safety baseline value , the grid cell is marked as a high-pressure area;

[0168] Trend calculation: for the high-pressure area, the time rate of change of the average osmotic pressure difference, i.e. the first derivative, is calculated using a sliding window:

[0169]

[0170] wherein is the average osmotic pressure difference of the grid cell at the current time, is the average osmotic pressure difference at the previous sampling time; is the system preset sampling time interval, for example 1 second, used to normalize the discrete pressure difference change to the time rate of change;

[0171] Zero-sample early warning: if the sensors in the high-pressure area do not trigger the conventional threshold alarm, and the rate of change of the area continuously exceeds the preset slope threshold , for example 0.5 / s, and the duration exceeds the time window , for example 10 seconds, the system determines that there is an atypical risk pattern, generating a zero-sample hidden danger early warning signal; this pattern means that although the physical value does not reach the alarm line, the entropy and cognitive difference of the data are rapidly increasing, indicating that unknown complex changes are occurring in the physical field underground;

[0172] This embodiment enables the system to identify potential disaster precursors in advance through monitoring the accumulation of information pressure, making the system able to identify potential disaster precursors in advance in the case of failure of traditional threshold alarms, zero-sample early warning, realizing the leap from rule-based alarm to situation awareness-based early warning, greatly improving the sensitivity and foresight of mine safety monitoring.

[0173] Example 6:

[0174] The method for updating the global model includes:

[0175] ​The cloud server unpacks the received entropy energy data capsule to parse out the original waveform data;

[0176] The original waveform data is used to incrementally train the current global model to obtain updated global model parameters;

[0177] The system calculates the change amount of the updated global model parameters and the previous version model parameters, and generates a model update instruction to issue the updated global model parameters to the edge computing node in response to the change amount being greater than a preset convergence threshold;

[0178] The method for dynamically adjusting the weight for subsequent osmotic pressure difference calculation includes:

[0179] The edge computing node receives the model update instruction to parse out the updated global model parameters, replaces the local model parameters with the updated global model parameters, and records the parameter update frequency;

[0180] If the parameter update frequency increases, the value of the second weighting coefficient is increased to increase the weight of the cognitive difference in the osmotic pressure difference calculation; if the parameter update frequency decreases, the value of the second weighting coefficient is decreased.

[0181] The embodiment details the closed-loop mechanism of cloud model updating and edge self-adaptive weight adjustment;

[0182] In the cloud, the cloud server unpacks the received entropy energy data capsule to parse out the original waveform data; the original waveform data is used to incrementally train the current global model to obtain updated global model parameters;

[0183] The system calculates the change amount of the updated global model parameters and the previous version model parameters, and generates a model update instruction to issue the updated global model parameters to the edge computing node in response to the change amount being greater than a preset convergence threshold;

[0184] In the edge, the edge computing node receives the model update instruction to parse out the updated global model parameters; the updated global model parameters are used to replace the local model parameters, and the parameter update frequency is recorded , for example, defined as the number of updates per unit time;

[0185] The system adjusts the second weighting coefficient based on the parameter update frequency , specifically using the following incremental adjustment formula:

[0186]

[0187] wherein, is a frequency normalization reference constant, for example, taking a value of 1 Hz, used to eliminate The physical dimensions of the function input items ensure the physical rigor of mathematical operations; The preset sensitivity adjustment coefficient, for example, 0.1, has its value strictly limited to a range of values. To ensure that In the extreme case where the function output is -1, the updated weight coefficients It remains positive throughout, maintaining the positive driving effect of model difference on osmotic pressure difference; The function is used to map the amplitude of frequency changes to The range is used to prevent weight oscillations;

[0188] when When the function value is positive, As the number of data points increases, the system is more inclined to upload data with cognitive differences in order to accelerate model convergence.

[0189] when When the function value is negative, As the value decreases, the system enters a steady-state maintenance mode.

[0190] This embodiment implements adaptive closed-loop control of the scheduling algorithm. The system can automatically adjust its learning desire based on the stability of the environment, upload more data and switch to energy-saving mode to reduce the balance between uploading. Updates are only sent when model parameters change significantly, avoiding the occupation of network resources by frequent downlink communication, while ensuring that the edge model can learn the latest downhole features in a timely manner.

[0191] Example 7:

[0192] Data collected from underground sensors in coal mines includes:

[0193] Microsecond-level time synchronization between underground sensors and edge computing nodes is achieved through PTP Precision Time Protocol or NTP Network Time Protocol to ensure that the timestamps of all collected data are based on a unified clock source; vibration waveform data and gas concentration time series data are collected in real time through vibration sensors and gas sensors deployed in coal mining faces and roadways.

[0194] The vibration waveform data and gas concentration time series data are denoised and preprocessed to remove invalid data segments caused by electromagnetic interference from the equipment.

[0195] This embodiment is a further specification of the data acquisition source in Embodiment 1;

[0196] The system collects vibration waveform data and gas concentration time series data in real time by deploying vibration sensors and gas sensors in the coal mining face and roadway.

[0197] The system carries out denoising preprocessing on the vibration waveform data and the gas concentration time sequence data, and eliminates invalid data segments generated due to electromagnetic interference of equipment; in this process, wavelet transform or Kalman filtering algorithm is used to remove high-frequency electromagnetic noise, so as to ensure that the effective physical signal enters the entropy energy data capsule;

[0198] The embodiment guarantees the quality of input data from the source, prevents noise data from being incorrectly identified as high-entropy data due to its randomness and occupying valuable transmission bandwidth, and improves the overall signal-to-noise ratio and scheduling accuracy of the system.

[0199] Embodiment 8:

[0200] Please refer to Figure 2 A coal mine data processing system based on cloud edge collaboration is used to implement the coal mine data processing method based on cloud edge collaboration as in any one of embodiments 1-7, comprising:

[0201] A data acquisition and encapsulation module is configured to acquire coal mine underground sensor data, and encapsulate the sensor data into a data packet containing a local information entropy parameter at an edge computing node, defined as an entropy energy data capsule;

[0202] A permeation model construction module is configured to construct a scheduling model based on a virtual permeation membrane logic, calculate a transmission priority value of the entropy energy data capsule relative to the cloud server based on the flow resistance parameter of the current network channel and the model parameter difference degree between the edge computing node and the cloud server, and define it as a permeation pressure difference;

[0203] A shunt execution module is configured to execute a data shunt strategy based on the permeation pressure difference: if the permeation pressure difference is greater than a preset threshold, the entropy energy data capsule is sent to the cloud server through the network; if the permeation pressure difference is less than or equal to the preset threshold, the entropy energy data capsule is executed for local compression storage and an index is generated, the process is defined as local crystallization storage, and a data fingerprint is extracted and synchronized to the cloud server;

[0204] A cloud feedback module is configured to update the global model according to the received entropy energy data capsule, and feed back the model parameter difference to the edge computing node, so as to dynamically adjust the weight coefficient when the edge computing node calculates the permeation pressure difference subsequently.

[0205] The embodiment provides a coal mine data processing system based on cloud edge collaboration, which relies on an edge gateway and a cloud server in a hardware architecture, and specifically includes the following modules in a software function;

[0206] A data acquisition and encapsulation module is configured to acquire coal mine underground sensor data, and encapsulate the sensor data into a data packet containing a local information entropy parameter at an edge computing node, defined as an entropy energy data capsule;

[0207] The permeation model construction module is configured to construct a scheduling model based on a virtual permeation membrane logic, calculate a transmission priority value of the entropy energy data capsule relative to the cloud server based on a flow resistance parameter of a current network channel and a model parameter difference degree between the edge computing node and the cloud server, and define the transmission priority value as a permeation pressure difference. As a core calculation unit of the system, the module is responsible for quantifying the transmission urgency of data.

[0208] The shunt execution module is configured to execute a data shunt strategy based on the permeation pressure difference.

[0209] In response to the permeation pressure difference being greater than a preset threshold, the entropy energy data capsule is sent to the cloud server through the network.

[0210] In response to the permeation pressure difference being less than or equal to the preset threshold, the entropy energy data capsule is locally compressed and stored and an index is generated, and a data fingerprint is extracted and synchronized to the cloud server.

[0211] The cloud feedback module is configured to update a global model according to the received entropy energy data capsule, and feed back a model parameter difference to the edge computing node, so that the edge computing node dynamically adjusts a weight coefficient when subsequently calculating the permeation pressure difference.

[0212] The present embodiment realizes decoupled deployment of the virtual permeation membrane logic on physical devices through modular design. The modules work cooperatively, so that the system can stably perform a data hierarchical scheduling task in a complex heterogeneous computing environment in a coal mine underground, and ensures the extreme utilization of software and hardware resources.

[0213] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A coal mine data processing method based on cloud-edge collaboration, characterized in that, Running in a heterogeneous system comprising edge computing nodes and cloud servers, the method includes: Data from underground sensors in a coal mine is collected, and the sensor data is encapsulated into a data packet containing local information entropy parameters at an edge computing node. The data packet is defined as an entropy energy data capsule. A scheduling model based on virtual osmotic membrane logic is constructed to obtain the flow resistance parameters of the current network channel and the cognitive difference of model parameters between the edge computing node and the cloud server. The transmission priority value of the entropy energy data capsule relative to the cloud server is calculated, and the transmission priority value is defined as the osmotic pressure difference. Execute a data diversion strategy based on the osmotic pressure difference: determine the relationship between the osmotic pressure difference and a preset threshold; if the osmotic pressure difference is greater than the preset threshold, send the entropy energy data capsule to the cloud server via the network; if the osmotic pressure difference is less than or equal to the preset threshold, perform local compression storage on the entropy energy data capsule and generate an index, define this process as local crystallization storage, and extract the data fingerprint in the index and synchronize it to the cloud server. The cloud server updates the global model based on the received entropy energy data capsules and feeds back the parameter differences between the global model and the local model to the edge computing nodes. The edge computing nodes dynamically adjust the weighting coefficients when calculating the osmotic pressure difference in subsequent calculations based on the parameter differences.

2. The coal mine data processing method based on cloud-edge collaboration according to claim 1, characterized in that, The data structure of the entropy energy data capsule includes raw waveform data, local information entropy, thermal attenuation coefficient, and edge computing power potential energy; The method of encapsulating entropy energy data into an entropy energy capsule containing local information entropy parameters includes: Obtain the original waveform data, calculate the data distribution sparsity value of the original waveform data, and mark the data distribution sparsity value as local information entropy; Obtain the difference between the timestamp of data acquisition and the current system time, calculate the attenuation factor that changes with time based on the difference, and mark the attenuation factor as the thermal attenuation coefficient; Obtain the current CPU idle rate of the edge computing node, calculate the reciprocal of the CPU idle rate, and mark the reciprocal as the edge computing power potential energy; The original waveform data, local information entropy, thermal attenuation coefficient, and edge computing potential energy are combined to generate the entropy energy data capsule.

3. The coal mine data processing method based on cloud-edge collaboration according to claim 2, characterized in that, The method for calculating the osmotic pressure difference of the entropy energy data capsule relative to the cloud server includes: Obtain the global model parameters fed back by the cloud server and the current model parameters of the edge computing node, calculate the gradient residual between the parameter vectors of the two, and mark the gradient residual as the cognitive difference. Obtain the current network channel bandwidth occupancy and packet loss rate, and calculate the network flow resistance parameters based on the bandwidth occupancy and packet loss rate; Obtain the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient; Multiply the local information entropy by the first weighting coefficient to obtain the entropy energy term; Multiply the cognitive difference by the second weighting coefficient to obtain the potential energy term; Multiply the network flow resistance parameters by the third weighting coefficient to obtain the resistance term; Calculate the sum of the entropy energy term and the potential energy term, then subtract the resistance term, and use the calculated result as the osmotic pressure difference.

4. The coal mine data processing method based on cloud-edge collaboration according to claim 3, characterized in that, The method for performing local crystallization storage of entropy energy data capsules includes: The original waveform data in the entropy energy data capsule is compressed using a lossless compression algorithm to generate compressed data blocks; The compressed data block is written into the non-volatile memory of the edge computing node; Perform a hash operation on the entropy energy data capsule to generate a unique identifier, and mark the unique identifier as a data fingerprint; Construct an index data packet containing the data fingerprint and the osmotic pressure difference, and send the index data packet to the cloud server.

5. The coal mine data processing method based on cloud-edge collaboration according to claim 4, characterized in that, The method further includes: The cloud server receives index data packets from multiple edge computing nodes; Parse the index data packet to obtain the osmotic pressure difference and sensor geographical location information corresponding to the data fingerprint; The sensor's geographic location information is mapped to the mine's spatial coordinate system, and the osmotic pressure difference is used as a thermal value to construct a pressure field distribution map of the entire mine. If the sensors in the high-pressure area do not trigger the conventional threshold alarm, but the average permeability differential in the area continues to rise, a zero-sample hazard warning signal is generated.

6. The coal mine data processing method based on cloud-edge collaboration according to claim 5, characterized in that, The method for updating the global model includes: The cloud server decapsulates the received entropy energy data capsule and extracts the original waveform data; The original waveform data is used to perform incremental training on the current global model to obtain the updated global model parameters; Calculate the change in the updated global model parameters compared to the previous version of the model parameters; If the change exceeds a preset convergence threshold, a model update instruction is generated, and the updated global model parameters are sent to the edge computing nodes.

7. The coal mine data processing method based on cloud-edge collaboration according to claim 6, characterized in that, The method for dynamically adjusting the weights used in calculating subsequent osmotic pressure differences includes: The edge computing node receives the model update instruction and parses out the updated global model parameters; Replace the local model parameters with the updated global model parameters and record the parameter update frequency; The second weighting coefficient is adjusted based on the parameter update frequency: if the parameter update frequency increases, the value of the second weighting coefficient is increased to increase the weight of cognitive differences in the calculation of osmotic pressure difference; if the parameter update frequency decreases, the value of the second weighting coefficient is decreased.

8. The coal mine data processing method based on cloud-edge collaboration according to claim 1, characterized in that, The collected data from underground sensors in the coal mine includes: Vibration and gas sensors deployed in coal mining faces and roadways are used to collect vibration waveform data and gas concentration time series data in real time. The vibration waveform data and gas concentration time series data are denoised and preprocessed to remove invalid data segments caused by electromagnetic interference from the equipment.

9. A cloud-edge collaborative coal mine data processing system, used to implement the cloud-edge collaborative coal mine data processing method as described in any one of claims 1-8, characterized in that, include: The data acquisition and encapsulation module is configured to acquire sensor data from underground coal mines and encapsulate the sensor data into a data packet containing local information entropy parameters at the edge computing node, which is defined as an entropy energy data capsule. The permeation model construction module is configured to construct a scheduling model based on virtual permeation membrane logic. Based on the current network channel flow resistance parameters and the cognitive difference of model parameters between edge computing nodes and cloud servers, the module calculates the transmission priority value of the entropy energy data capsule relative to the cloud server, which is defined as the permeation pressure difference. The data diversion execution module is configured to execute a data diversion strategy based on the osmotic pressure difference: if the osmotic pressure difference is greater than a preset threshold, the entropy energy data capsule is sent to the cloud server via the network; If the osmotic pressure difference is less than or equal to a preset threshold, the entropy energy data capsule is locally compressed and stored and an index is generated, defined as local crystallized storage, and the data fingerprint is extracted and synchronized to the cloud server. The cloud feedback module is configured to update the global model based on the received entropy energy data capsules and feed back the model parameter differences to the edge computing nodes so that the edge computing nodes can dynamically adjust the weight coefficients when calculating the osmotic pressure difference in the future.

Images