Intelligent document positioning method and system based on finite state constraints

Abstract

The application belongs to the technical field of network data processing and digital information transmission, and particularly relates to an intelligent official document positioning method and system based on a limited state constraint, which comprises probability credibility judgment, circulation state fusion and circulation state judgment. The application solves the problem that the existing prior art has a high misjudgment rate of an object positioning result due to the influence of self-positioning errors and internal and external interference factors, and has the beneficial technical effects of overcoming errors caused by self errors and external interference, greatly reducing the misjudgment rate of the positioning result, and improving the object positioning accuracy.

Description

Technical Field

[0001] This invention belongs to the field of network data processing and digital information transmission technology, and particularly relates to an intelligent document location method and system based on finite state constraints. Background Technology

[0002] In the daily management work of government agencies, universities, research institutes, and large enterprises, important official documents often need to be located and managed. The existing document management has the following significant problems: 1. Due to various reasons, phenomena such as "document travel" and "snail-like operation" often occur; 2. For some reason, the documents are in the office building but cannot be found; 3. Sometimes, due to work negligence, confidential documents may be taken to public areas, posing a risk of security leaks.

[0003] Currently, passive positioning using Bluetooth gateways is a common method in indoor positioning. A Bluetooth gateway is a device that can receive and transmit Bluetooth signals. By receiving signals from surrounding Bluetooth devices, it can determine the location information of these devices. This method has low power consumption and cost, and therefore is widely used in indoor positioning and other scenarios. However, the accuracy of Bluetooth gateway positioning is affected by a variety of factors, including signal interference, transmission distance, and wall obstruction. At the same time, using multiple Bluetooth gateways for collaborative positioning can improve positioning accuracy and reliability. Currently, Bluetooth gateway positioning technology has been applied in fields such as indoor navigation, personnel positioning, and object tracking.

[0004] In existing indoor object positioning methods, the positioning accuracy and stability of Bluetooth positioning are easily affected by the number and density of installed devices and external interference. When determining which room a file is located in, deviations are prone to occur, causing the positioning results to jump between adjacent rooms on the same floor or between the same room number on different floors.

[0005] In summary, existing technologies suffer from a high misjudgment rate in object positioning results due to their own positioning errors and the influence of internal and external interference factors. Summary of the Invention

[0006] This invention provides an intelligent document location method and system based on finite state constraints to solve the problem mentioned in the background art, where the existing technology has a high misjudgment rate of item location results due to its own positioning error and the influence of internal and external interference factors.

[0007] The technical problem solved by this invention is achieved by the following technical solution:

[0008] Intelligent document location methods based on finite state constraints include:

[0009] Probability Confidentiality Assessment: Based on the probability of a file's existence in a room determined by the location prediction model, a probability difference is calculated through probability comparison. This probability difference is then compared with a probability threshold to determine the file's credibility. If the file's credibility meets the threshold, the room affiliation of the file can be immediately determined; otherwise, the following steps are continued:

[0010] State transition fusion: Based on statistical calculations of the state transition process, the probability of file state transition is determined. Based on the file state transition probability, the probability of state transition fusion is determined through state transition fusion calculation.

[0011] Flow status determination: Based on the fusion probability of flow status, the room to which the file flows is assigned is determined through probability comparison analysis, and the room with the higher fusion probability of the file flow status is selected as the result room.

[0012] further:

[0013] The probability confidence assessment also includes probability comparison calculation, which includes:

[0014] based on:

[0015] The probability M1 of the maximum value file corresponding to the first room exists, and;

[0016] The probability M2 of the second largest value file corresponding to the second room exists;

[0017] The probability difference is determined by a probability comparison function;

[0018] The probability comparison function is:

[0019] E = (M1 - M2) / M2;

[0020] E represents the probability dissimilarity level;

[0021] M1 represents the probability of the maximum value file existing in all rooms corresponding to the first room;

[0022] M2 represents the probability of the second largest value file existing in all rooms corresponding to the second room.

[0023] further:

[0024] The probability confidence assessment also includes probability threshold comparison, which includes:

[0025] Based on the probability difference, a threshold comparison function is used. If the probability difference is greater than the probability comparison threshold, the probability credibility meets the standard; otherwise, the probability credibility does not meet the standard.

[0026] The threshold comparison function is:

[0027]

[0028] E represents the probability dissimilarity level;

[0029] C0 is the probability comparison threshold.

[0030] further:

[0031] The state transition fusion probability also includes statistical calculation of the state transition process, which includes:

[0032] based on:

[0033] For all states, the sum of the cumulative number of state transitions that flow out of that state within a given period;

[0034] The cumulative number of times a process transitions to a specific state during all these state transitions;

[0035] The file transition probability is ultimately determined by the state transition probability function;

[0036] The state transition probability function is:

[0037] P = N1N;

[0038] P represents the state transition probability;

[0039] N represents the cumulative number of all processes that have flowed out of a certain state, as previously recorded.

[0040] N1 represents the total number of processes that transition from this state to a specific state within a given period.

[0041] further:

[0042] The state transition fusion probability further includes state transition fusion calculation, which includes: based on:

[0043] The probability of a file transitioning from the previous state to the first room, and;

[0044] The probability of a file transitioning from the previous state to the second room;

[0045] Determine using the state fusion probability function:

[0046] The corresponding file transfer status fusion probability to the first room, and;

[0047] The probability of merging the corresponding file flow status when it is transferred to the second room;

[0048]

[0049] The POI is the probability of a file being transferred to the first room.

[0050] The POJ represents the probability of a file being transferred to the second room.

[0051] M1 represents the probability of the existence of the maximum value file corresponding to the first room.

[0052] M2 represents the probability of the existence of the second largest value file corresponding to the second room.

[0053] The PI is the probability of fusion of the file's transfer status to the first room;

[0054] PJ represents the probability of file transfer to the second room.

[0055] further:

[0056] The flow status determination also includes probability comparison analysis, which includes:

[0057] based on:

[0058] The probability of file transfer status fusion when it is transferred to the first room, and;

[0059] The probability of merging the flow status of files when they are transferred to the second room;

[0060] The accuracy of file transfer is determined by comparing the transfer status.

[0061]

[0062] The PI is the probability of fusion of the file's transfer status to the first room;

[0063] PJ represents the probability of file transfer to the second room.

[0064] further:

[0065] The flow state fusion also includes flow state amplification calculation, which includes:

[0066] based on:

[0067] The probability of file transfer status fusion when it is transferred to the first room, and;

[0068] The probability of merging the flow status of files when they are transferred to the second room;

[0069] The probability of fusion of flowing states is amplified by the flowing state amplification function;

[0070] The flow state amplification function includes:

[0071]

[0072] The POI is the probability of a file being transferred to the first room state;

[0073] The POJ represents the probability of a file being transferred to the second room state.

[0074] PI' is the amplified fusion probability of the flow state;

[0075] PJ' is the amplified fusion probability of the flow state;

[0076] The amplified probability comparison analysis includes:

[0077] based on:

[0078] The magnified fusion probability of the file's flow state after it is transferred to the first room, and;

[0079] The probability of fusion of the magnified flow state of the file when it is transferred to the second room;

[0080] The specific room for file flow is determined by a comparison function that amplifies the flow state; the room with a higher probability of file flow state fusion is selected as the result room.

[0081]

[0082] PI' is the amplified fusion probability of the flow state;

[0083] PJ' represents the amplified fusion probability of the flow state.

[0084] further:

[0085] Before the probability credibility judgment, the document existence probability acquisition is also included. The document existence probability acquisition includes: Bluetooth data acquisition, data feature extraction, positioning model training, and document existence probability calculation.

[0086] Bluetooth data acquisition: Acquire room calibration data, which includes the corresponding Bluetooth wireless signal information of each reference node and acquisition node, as well as the corresponding room location information;

[0087] Data feature extraction: Based on the collected Bluetooth data, the raw Bluetooth data is converted into corresponding signal feature vectors through information feature extraction;

[0088] Location model training: Based on the room location information and the signal feature vectors mentioned above, a high-speed learning machine model is used for training to obtain a location prediction model.

[0089] Based on the input signal feature vector and the corresponding positioning prediction model, the probability of document presence in each room is obtained through the calculation method of a high-speed learning machine.

[0090] further:

[0091] The location prediction model is as follows:

[0092] T = H × B;

[0093] H is the hidden layer output matrix;

[0094] B is the output weight matrix;

[0095] T represents the probability of the existence of the corresponding file in each room.

[0096] Meanwhile, the present invention also provides an intelligent document location system based on finite state constraints, including: an intelligent document location platform that implements the above-mentioned intelligent document location method, wherein the intelligent document location platform includes a probability credibility judgment module, a circulation state fusion module, and a circulation state judgment module;

[0097] The probability credibility judgment module is used to: determine the probability of a file's existence in a room based on the location prediction model; calculate the probability comparison difference through probability comparison; determine the credibility of the file's existence by comparing it with a probability threshold based on the probability comparison difference; and determine whether the credibility of the file's existence meets the standard. If the credibility of the file's existence probability meets the standard, the file's ownership is directly determined; otherwise, subsequent steps are continued.

[0098] The flow state fusion module is used to: determine the file state flow probability based on statistical calculations of the state flow process; and determine the flow state fusion probability through state transition fusion calculations based on the file state flow probability.

[0099] The circulation status judgment module is used to: determine the room affiliation of a file circulation based on the circulation status fusion probability through probability comparison analysis, and select the room with the higher circulation status fusion probability as the result room.

[0100] Beneficial technical effects:

[0101] This patent employs a probabilistic credibility judgment: based on the probability of a file's existence in a room determined by a location prediction model, a probability difference is calculated through probability comparison. Based on this probability difference, the credibility of the file's existence is determined by comparing it with a probability threshold. If the credibility of the file's existence probability meets the threshold, the room affiliation of the file can be immediately determined; otherwise, the following steps are continued: Flow state fusion: based on statistical calculations of the state flow process, the probability of file state flow is determined. Based on the file state flow probability, a state transition fusion calculation is performed to determine the flow state fusion probability; Flow state judgment: based on the flow state fusion probability, through probability comparison analysis, the room affiliation of the file flow is determined, and the room with the higher file flow state fusion probability is selected as the result room. Since firstly, as... Figure 4As shown, this invention uses Bluetooth gateway positioning technology for positioning. 1. A Bluetooth beacon is fixed on the document folder, emitting a standard Bluetooth signal; 2. A Bluetooth gateway (with different appearances) installed in the environment collects the Bluetooth signal and sends it to a Bluetooth positioning engine located in the cloud. The Bluetooth positioning engine is located on a cloud-based network server and includes specific calculation methods and supporting software tools. Simultaneously, Bluetooth beacons are installed on all documents, and Bluetooth gateways are installed in locations requiring positioning. The gateways locate the Bluetooth beacons (documents), determine their room number, and further determine their dwell time, ultimately achieving intelligent and secure management of documents during circulation. Due to the instability of the Bluetooth signal itself and environmental interference, the result may fluctuate, i.e., the positioning result may jump between adjacent rooms on the same floor or between rooms with the same number on different floors. The so-called document circulation state means that a document will only exist in a limited number of states during the entire circulation process, as shown in the figure: State 1), located in the management office; State 2), located in the secretary's office; State 3), located in the leader's office; State 4), ... 4) Located in the storage room, where: State 1), located in the management office: the first place where the document is logged into the management system; State 2), located in the secretary's office: the document will enter and exit the secretary's office multiple times; State 3), located in the leader's office: generally transferred from the secretary's office to the leader's office for processing, and then enters the secretary's office again; State 4), located in the storage room: after the case is closed, the document will be stored in the storage room; there are certain probabilistic flow relationships between the four states. In order to solve the above problems, in this case, optimization methods with document flow state constraints, such as probability credibility judgment, flow state fusion, and flow state judgment, are used. Specifically: based on the document existence probability of the room determined by the positioning prediction model, the maximum and second largest values ​​of the document existence probability of all rooms are selected. By comparing the probability thresholds, a new probability is calculated by fusing the document existence probability and the document flow probability. Then, the size of the fused probability is compared to finally determine the room to which the document flow belongs. By fusing the existence probability and the flow probability, the errors caused by its own error and external interference are overcome, the misjudgment rate of the positioning result is greatly reduced, and the accuracy of item positioning is improved. Attached Figure Description

[0102] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0103] Figure 1 This is the overall flowchart of the document location method of the present invention;

[0104] Figure 2 This is a flowchart illustrating the document location method of the present invention.

[0105] Figure 3 This is an application flowchart of the document location method of the present invention;

[0106] Figure 4 This is a Bluetooth gateway positioning structure topology diagram of the document positioning method of the present invention;

[0107] Figure 5 This is a schematic diagram of the document flow status of the document location method of the present invention;

[0108] Figure 6 This is a diagram of the Bluetooth positioning machine learning neural network structure of the document positioning method of this invention. Detailed Implementation

[0109] The present invention will be further described below with reference to the accompanying drawings:

[0110] In the picture:

[0111] S101 - Probability and credibility assessment;

[0112] S102 - Fusion of Flowing State;

[0113] S103 - Flow status judgment;

[0114] S2001 - Based on the probability of a file's existence in a room determined by the location prediction model, the probability difference is calculated through probability comparison. Based on the probability difference, the credibility of the file's existence is determined by comparing it with a probability threshold. If the credibility of the file's existence probability meets the standard, the room ownership of the file can be immediately determined; otherwise, the following steps are continued.

[0115] S2002 - Based on statistical calculations of the state transition process, determine the probability of file state transition; based on the probability of file state transition, determine the probability of state transition fusion through state transition fusion calculation.

[0116] S2003 - Based on the fusion probability of file circulation status, the room affiliation of the file circulation is determined through probability comparison analysis, and the room with the higher file circulation status fusion probability is selected as the result room.

[0117] Example:

[0118] This embodiment: as follows Figure 1 , 2 As shown, the intelligent document location method based on finite state constraints includes:

[0119] Probability Confidentiality Judgment S101: Based on the probability of a file's existence in a room determined by the location prediction model, the probability difference is calculated through probability comparison. Based on the probability difference, the file's existence credibility is determined by comparing it with a probability threshold. If the file's existence probability credibility meets the standard, the room affiliation of the file can be immediately determined; otherwise, the following steps S2001 are continued:

[0120] S102: Based on statistical calculations of the state transition process, determine the file state transition probability. Based on the file state transition probability, determine the state transition fusion probability S2002 through state transition fusion calculation.

[0121] Flow status judgment S103: Based on the flow status fusion probability, the room affiliation of the file flow is determined through probability comparison analysis, and the room with the higher file flow status fusion probability is selected as the result room S2003.

[0122] Because a probability-based reliability judgment is used: based on the probability of a file's existence in a room determined by the location prediction model, a probability difference is calculated through probability comparison. Based on this probability difference, the reliability of the file's existence is determined by comparing it with a probability threshold. If the reliability of the file's existence probability meets the standard, the room affiliation of the file can be immediately determined; otherwise, the following steps are continued: Flow state fusion: based on statistical calculations of the state flow process, the probability of file state flow is determined. Based on the file state flow probability, a state transition fusion calculation is performed to determine the flow state fusion probability; Flow state judgment: based on the flow state fusion probability, a probability comparison analysis is performed to determine the room affiliation of the file flow, and the room with the higher file flow state fusion probability is selected as the result room. Since firstly, as... Figure 4 As shown, this invention uses Bluetooth gateway positioning technology for positioning. 1. A Bluetooth beacon is fixed on the document folder, emitting a standard Bluetooth signal; 2. A Bluetooth gateway (with different appearances) installed in the environment collects the Bluetooth signal and sends it to a Bluetooth positioning engine located in the cloud. The Bluetooth positioning engine is located on a cloud-based network server and includes specific calculation methods and supporting software tools. Simultaneously, Bluetooth beacons are installed on all documents, and Bluetooth gateways are installed in locations requiring positioning. The gateways locate the Bluetooth beacons (documents), determine their room number, and further determine their dwell time, ultimately achieving intelligent and secure management of documents during circulation. Due to the instability of the Bluetooth signal itself and environmental interference, the result may fluctuate, i.e., the positioning result may jump between adjacent rooms on the same floor or between rooms with the same room number on different floors. The so-called document circulation state means that a document will only exist in a limited number of states throughout the entire circulation process, such as... Figure 5As shown, the states are: State 1), located in the management office; State 2), located in the secretary's office; State 3), located in the leader's office; State 4), located in the storage room. Specifically, State 1), located in the management office: the document is first logged into the management system; State 2), located in the secretary's office: the document will enter and exit the secretary's office multiple times; State 3), located in the leader's office: generally transferred from the secretary's office to the leader's office for processing, and then back to the secretary's office; State 4), located in the storage room: after the case is closed, the document will be stored in the storage room. There is a certain probability of flow between these four states. To solve the above problems, in this case, through... Optimization methods with file flow status constraints, such as probability reliability judgment, flow status fusion, and flow status judgment, are as follows: Based on the file existence probability of a room determined by the positioning prediction model, the maximum and second largest file existence probabilities of all rooms are selected. By comparing the probability thresholds, a new probability is calculated by fusing the file existence probability and the file flow probability. The magnitude of the fused probability is then compared to determine the room to which the file flow belongs. By fusing the existence probability and the flow probability, errors caused by its own error and external interference are overcome, greatly reducing the misjudgment rate of positioning results and improving the accuracy of item positioning.

[0123] like Figure 3 As shown:

[0124] The probability confidence judgment S101 further includes probability comparison calculation, which includes:

[0125] based on:

[0126] The probability M1 of the maximum value file corresponding to the first room exists, and;

[0127] The probability M2 of the second largest value file corresponding to the second room exists;

[0128] The probability difference is determined by a probability comparison function;

[0129] The probability comparison function is:

[0130] E = (M1 - M2) / M2;

[0131] E represents the probability dissimilarity level;

[0132] M1 represents the probability of the maximum value file existing in all rooms corresponding to the first room;

[0133] M2 represents the probability of the second largest value file existing in all rooms corresponding to the second room.

[0134] Since the probability difference is determined by using the probability M1 of the maximum value file corresponding to the first room and the probability M2 of the second maximum value file corresponding to the second room, a probability comparison function is used. The probability comparison function is:

[0135] E = (M1 - M2) / M2;

[0136] Based on the data result T of the prediction model, the output data is judged by comparing the maximum value M1 with the second largest value M2. If (M1-M2) / M2>=C0, it means that the positioning result is reliable enough, and the room corresponding to the maximum value M1 is directly taken as the positioning result. The above scheme is the algorithm and calculation function for probability difference judgment.

[0137] The probability confidence judgment S101 further includes probability threshold comparison, which includes:

[0138] Based on the probability difference, a threshold comparison function is used. If the probability difference is greater than the probability comparison threshold, the probability credibility meets the standard; otherwise, the probability credibility does not meet the standard.

[0139] The threshold comparison function is:

[0140]

[0141] E represents the probability dissimilarity level;

[0142] C0 is the probability comparison threshold.

[0143] Since the probability confidence assessment also includes probability threshold comparison, the probability threshold comparison includes:

[0144] Based on the probability difference, a threshold comparison function is used. If the probability difference is greater than the probability comparison threshold, the probability reliability meets the standard; otherwise, the probability reliability does not meet the standard. The threshold comparison function is:

[0145]

[0146] Since C0 is a confidence parameter, which is related to system performance, it is generally taken as 0.5. The above scheme is a threshold judgment for probability difference.

[0147] The state transition fusion also includes statistical calculation of the state transition process, which includes:

[0148] based on:

[0149] For all states, the sum of the cumulative number of state transitions that flow out of that state within a given period;

[0150] The cumulative number of times a process transitions to a specific state during all these state transitions;

[0151] The file transition probability is ultimately determined by the state transition probability function;

[0152] The state transition probability function is:

[0153] P = N1 / N;

[0154] P represents the state transition probability;

[0155] N represents the cumulative number of all processes that have flowed out of a certain state, as previously recorded.

[0156] N1 represents the total number of processes that transition from this state to a specific state within a given period.

[0157] Since the state transition fusion probability also includes statistical calculation of the state transition process, the statistical calculation of the state transition process includes: based on all states, the cumulative number of state transitions from each state within a certain period and the cumulative number of transitions to a specific state during all these state transitions; the file transition probability is finally determined through the state transition probability function; the state transition probability function is:

[0158] P = N1 / N;

[0159] Since the probability P between each circulation state is calculated from the actual document data during circulation, i.e., P_process = total number of circulation processes / total number of circulations, each of the above four states can be identified by multiple rooms. Furthermore, each room can only have one of the four state identifiers. Because the document processing state of multiple rooms is the same, for example, if a secretary's office has n rooms, then all n rooms represent the state "Secretary's Office".

[0160] like Figure 5 As shown, the prior probability P between each flow state is calculated from the statistical data of the actual file during the flow, that is:

[0161] P(process) = Total number of times this process is executed / Total number of times the process is executed;

[0162] For example, the calculations for P2 and P4 in the diagram are shown in the table below:

[0163] Table 1: Number of times data was collected during the one-month observation period

[0164]

[0165] The state transition fusion S102 further includes state transition fusion calculation, which includes:

[0166] based on:

[0167] The probability of a file transitioning from the previous state to the first room, and;

[0168] the transition probability of the file transitioning from the previous state to the second room;

[0169] determine by using a state fusion probability function:

[0170] the corresponding transition state fusion probability of the file transitioning to the first room, and the corresponding transition state fusion probability of the file transitioning to the second room;

[0171] The state fusion probability function is:

[0172]

[0173] The POI is the state transition probability of the file transitioning to the first room;

[0174] The POJ is the state transition probability of the file transitioning to the second room;

[0175] The M1 is the maximum file existence probability corresponding to the first room;

[0176] The M2 is the second-maximum file existence probability corresponding to the second room;

[0177] The PI is the transition state fusion probability of the file transitioning to the first room;

[0178] The PJ is the transition state fusion probability of the file transitioning to the second room.

[0179] Since the adopted transition state fusion probability further includes state transition fusion calculation, the state transition fusion calculation comprises: based on the transition probability of the file transitioning from the previous state to the first room and the transition probability of the file transitioning from the previous state to the second room; determining, by a state fusion probability function, the corresponding transition state fusion probability of the file transitioning to the first room and the corresponding transition state fusion probability of the file transitioning to the second room; the state fusion probability function is:

[0180]

[0181] Since if (M1-M2) / M2 < C0, the state transition fusion calculation needs to be performed: assuming that the room corresponding to M1 has a corresponding state I state, the room corresponding to M2 has a corresponding state J state, and the previous state of file transition is state O, the probability of the state transition process OI is POI, and the probability of the state transition process OJ is POJ, two new probabilities can be calculated, that is, PI = POI × M1, PJ = POJ × M2, and the new probability obtained by fusing the file existence probability and the file transition probability provides a new basis for final positioning.

[0182] The transition state determination S103 further comprises probability comparative analysis, and the probability comparative analysis comprises:

[0183] Based on:

[0184] the fusion probability of the transfer state of the file transferred to the first room, and the fusion probability of the transfer state of the file transferred to the second room;

[0185] determining the file transfer accuracy through a transfer state comparison function;

[0186]

[0187] wherein PI is the fusion probability of the transfer state of the file transferred to the first room;

[0188] wherein PJ is the fusion probability of the transfer state of the file transferred to the second room.

[0189] Since the fusion probability of the transfer state of the file transferred to the first room and the fusion probability of the transfer state of the file transferred to the second room are adopted; the file transfer accuracy is determined through the transfer state comparison function; the transfer state comparison function is:

[0190]

[0191] Because if PI>=PJ, it is determined that the new room is the room corresponding to M1, otherwise if PI<PJ, it is determined that the new room is the room corresponding to M2, and the room where the file is located is determined more accurately through the above determination.

[0192] The transfer state fusion S102 further includes a transfer state amplification calculation, and the transfer state amplification calculation comprises:

[0193] Based on:

[0194] the fusion probability of the transfer state of the file transferred to the first room, and the fusion probability of the transfer state of the file transferred to the second room;

[0195] amplifying the fusion probability of the transfer state through a transfer state amplification function;

[0196] The transfer state amplification function comprises:

[0197]

[0198] wherein POI is the transfer probability of the state where the file is transferred to the first room;

[0199] wherein POJ is the transfer probability of the state where the file is transferred to the second room;

[0200] wherein PI' is the amplified fusion probability of the transfer state;

[0201] wherein PJ' is the amplified fusion probability of the transfer state;

[0202] The amplified probability comparison analysis includes:

[0203] based on:

[0204] The fusion probability of the magnified flow state of the file when it is transferred to the first room, and the fusion probability of the magnified flow state of the file when it is transferred to the second room;

[0205] The specific room for file flow is determined by a comparison function that amplifies the flow state; the room with a higher probability of file flow state fusion is selected as the result room.

[0206]

[0207] PI' is the amplified fusion probability of the flow state;

[0208] PJ' represents the amplified fusion probability of the flow state.

[0209] Because it employs the fusion probability of the file flow state from the first room to the second room and the fusion probability of the file flow state from the second room; the fusion probability of the flow state is amplified through a flow state amplification function; the flow state amplification function includes:

[0210]

[0211] Since the values ​​of PI and PJ can already be used to determine whether the current file is in room M1 (state I) or room M2 (state J), ​​a new metric method was invented to further improve the accuracy of the results. This method gives a greater advantage to rooms with higher transition probabilities when the machine model's judgment results are similar; that is, PI' = f(POI) * M1, PJ' = f(POJ) * M2; where the function f(x) is expressed as: f(x) = e x This effectively amplifies the input x and achieves good results.

[0212] Simultaneously, based on the amplified flow state fusion probability S102 of the file flow to the first room and the amplified flow state fusion probability S102 of the file flow to the second room, the specific room for file flow is determined through the flow state amplification comparison function; the room with the higher file flow state fusion probability is selected as its result room.

[0213]

[0214] The above method is based on the judgment of amplified fusion probability, which makes the judgment more accurate and thus improves the positioning accuracy.

[0215] Before the probability credibility judgment S101, the document existence probability acquisition is also included. The document existence probability acquisition includes: Bluetooth data acquisition, data feature extraction, positioning model training, and document existence probability calculation.

[0216] Bluetooth data acquisition: Acquire room calibration data, which includes the corresponding Bluetooth wireless signal information of each reference node and acquisition node, as well as the corresponding room location information;

[0217] Data feature extraction: Based on the collected Bluetooth data, the raw Bluetooth data is converted into corresponding signal feature vectors through information feature extraction;

[0218] Location model training: Based on the room location information and the signal feature vectors mentioned above, a high-speed learning machine model is used for training to obtain a location prediction model.

[0219] Based on the input signal feature vector and the corresponding positioning prediction model, the probability of document presence in each room is obtained through the calculation method of a high-speed learning machine.

[0220] Before using the aforementioned probability credibility judgment, the process also includes obtaining the file existence probability, which comprises: Bluetooth data acquisition, data feature extraction, positioning model training, and file existence probability calculation; Bluetooth data acquisition: acquiring room calibration data, which includes the corresponding Bluetooth wireless signal information of each reference node and acquisition node, as well as the corresponding room location information; data feature extraction: based on the acquired Bluetooth data, converting the raw Bluetooth data into corresponding signal feature vectors through information feature extraction; positioning model training: based on the aforementioned room location information and the signal feature vectors, using a high-speed learning machine model for training and learning to obtain a positioning prediction model; based on the input signal feature vectors and the corresponding positioning prediction model, obtaining the file existence probability of each room through the calculation method of the high-speed learning machine;

[0221] Since the core positioning algorithm of this invention adopts a machine learning framework, it is specifically divided into two stages. The first stage is the offline training stage, which typically includes three steps: Bluetooth data acquisition, feature extraction, and positioning model training. Bluetooth data acquisition usually refers to the acquisition of room calibration data. Room calibration data means that during the acquisition process, not only can the Bluetooth wireless signal information parameters of each reference point (RP) or acquisition node be obtained, but also the room location information corresponding to the acquisition of this set of Bluetooth wireless signals is known. After obtaining the data, feature extraction is performed to transform the Bluetooth wireless signal vector into a feature vector BV; then, these feature vectors and machine learning method F are used to train the positioning model LM. The second stage is the online positioning stage, which involves acquiring Bluetooth wireless signals at the test location and obtaining feature vector X using the same feature extraction method as the training data; this feature vector X is then input into the positioning model LM to calculate the user's actual location Y. In both stages, the feature extraction method used in this invention is to directly assemble the Bluetooth beacon signal strength (RSSI) acquired by each Bluetooth gateway at the same time into a feature vector in a fixed order. For some gateways that did not acquire signal strength at that time, the feature vector is not directly extracted. If , then 0 is used instead; for example: if a total of 10 gateways are installed in a certain site, and at a certain time t0, the signal strength values ​​collected by each gateway from the Bluetooth beacon emitted by the folder to be located are -50, -66, -98, none, none, -75, -77, -90, -89, -80, then the feature vector BV is (-50, -66, -98, 0, 0, -75, -77, -90, -89, -80). The machine learning method used in this invention can adopt existing technologies such as vector regression. Preferably, in this embodiment, the training process of this invention is as follows: for a set of collected input Bluetooth feature vectors x∈R n (where n is the number of gateways in the site) and an SLFN with L hidden layer nodes, its output form is defined as follows:

[0222]

[0223] Among them, a i =[a i1 ,a i2 ,L,a in ] T b represents the weight between the input node and the i-th hidden layer node. i G(a) represents the bias parameter of the i-th hidden layer node. i ,b i x) is the output of the i-th hidden layer node, β = [β i1 ,β i2 ,L,β im ] TThis represents the weight connecting the i-th hidden node and the output node T(t1, t2, ..., tm).

[0224] like Figure 6 As shown, T(t1, t2, ..., tm) represents the probability that the Bluetooth beacon exists in any of the m rooms in the venue. The parameter T serves as input during the first stage of training the machine learning model, representing the calibration results. In the second stage of prediction, T is the output; the larger the output value, the more likely the file is in that room. For example, if there are 15 rooms involved in the file transfer process, then if the file is in the first room, the value of T is (1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0); if the file is in the second room, the value of vector T is (0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), and so on. If T is the output result obtained by the model prediction calculation of a certain input Bluetooth signal vector, such as T = (0.5, 0.02, 0.005, 0.2, 0.06, 0.11, 0.23, 0.008, 0.03, 0.19, 0.21, 0.05, 0.07, 0.01, 0.03), then it can be determined that the current file is located in room 1.

[0225] The location prediction model is as follows:

[0226] T = H × B;

[0227] H is the hidden layer output matrix;

[0228] B is the output weight matrix;

[0229] T represents the probability of the existence of the corresponding file in each room.

[0230] The location prediction model used is as follows:

[0231] T = H × B;

[0232] As shown in the figure, the activation function of the hidden layer nodes in the model training network of this invention is selected as g(x). Typically, the activation function g(x) is chosen from functions such as "Sigmoid", "ReLU", and "RBF". Therefore:

[0233] G(a i ,b i ,x)=g(a i ·x+b i ),a i ∈R n ,b i ∈R

[0234] For the input N training sample datasets {(x j ,t j )|j=1,2,…,N},x j =[x j1 ,x j2 ,…,x jn ] T Among them, x j ∈R n t represents the feature vector of the input Bluetooth signal. j ∈R n Let the probability of a file existing in a specific room be represented by the sample data. Then:

[0235] G(a i ,b i ,x j )=g(a i ·x j +b i ),a i ∈R n ,b i ∈R,j=1,2,L,N

[0236] In summary, this can be represented in matrix form as follows:

[0237]

[0238]

[0239] According to Equation 1, we can obtain

[0240] Hβ=T

[0241] This formula indicates that the constructed SLFN will be trained with zero error; H represents the hidden layer output matrix, β represents the output weight matrix, and T represents the output room number matrix. Traditional single-hidden-layer neural networks use gradient descent to learn the input weight matrix, hidden node biases, and output weight matrix. However, gradient descent is prone to vanishing gradients or local optima, and the iterative optimization process leads to slow learning speed. Compared to traditional single-hidden-layer neural network learning methods, the high-speed learning machine trains the input weight matrix a... i and bias b i Random values ​​are assigned, requiring no adjustment during training; only the network output weight matrix β needs to be solved. Therefore, H and T are known, so β ​​is the parameter to be solved. The above equation can be regarded as a linear system, and solving the above equation can be achieved by solving the minimum mean square error of the system, which is equivalent to solving equation (6):

[0242]

[0243] Solving the above equation using the least squares method yields:

[0244]

[0245] in, It is the generalized inverse matrix of H.

[0246] at the same time:

[0247] In this embodiment, the process of the high-speed learning machine is as follows:

[0248] Step 1: Collect a batch of Bluetooth training data {(x j ,t j )|j=1,2,…,N},x j =[x j1 ,x j2 ,…,x jn ] T ;

[0249] Step 2: Randomly set the input weight matrix a i and bias parameter b i Determine the number of hidden layer nodes L, and set the activation function g(x);

[0250] Step 3: Calculate the hidden layer output weight matrix H;

[0251] Step 4: According to the formula T calculates the output weight matrix β of the network;

[0252] Step 5: Calculate the final output value Hβ of the network using β;

[0253] According to the formula Hβ=T, we know that once we obtain the output value T(t1, t2, ..., tm) of the machine learning model, we can determine the i-th room (where ti is the largest) of the m rooms based on the rules.

[0254] The method of the present invention achieved beneficial results in a specific implementation site with a total of 25 room detection points. During a one-month monitoring of the circulation status, after adopting the status constraint scheme, the judgment accuracy increased from 92.3% to 98.5%.

[0255] Total number of observations for document flow events 985 times 100％ Only the number of times the rapid learning machine correctly predicted. 910 times 92.3％ The number of times the method of this invention can be correctly predicted 970 times 98.5％ Improve results 60 6.2％

[0256] The intelligent document location system based on finite state constraints includes: an intelligent document location platform that implements the above-mentioned intelligent document location method, wherein the intelligent document location platform includes a probability credibility judgment module, a circulation state fusion module, and a circulation state judgment module;

[0257] The probability credibility judgment module is used to: determine the probability of a file's existence in a room based on the location prediction model, obtain the probability comparison difference through probability comparison calculation, determine the credibility of the file's existence based on the probability comparison difference and compare it with a probability threshold, and determine whether the credibility of the file's existence meets the standard. If the credibility of the file's existence probability meets the standard, the file's ownership is directly determined; if it does not meet the standard, the subsequent steps S2001 are continued.

[0258] The flow state fusion module is used to: determine the file state flow probability based on statistical calculations of the state flow process; and determine the flow state fusion probability S2002 based on the file state flow probability through state transition fusion calculations.

[0259] The circulation status judgment module is used to: determine the room affiliation of the file circulation based on the circulation status fusion probability through probability comparison analysis, and select the room with the higher circulation status fusion probability as its result room S2003.

[0260] Meanwhile, the present invention also provides an intelligent document location system based on finite state constraints, which includes an intelligent document location platform of the intelligent document location method. The intelligent document location platform includes a probability credibility judgment module, a circulation state fusion module, and a circulation state judgment module. Through the use of the above platform and corresponding modules, an application platform is provided for the intelligent document location method, thereby improving the practicality of the application of the intelligent document location method.

[0261] Working principle:

[0262] This patent uses probability reliability judgment: based on the probability of a file's existence in a room determined by a location prediction model, probability difference is calculated through probability comparison. Based on the probability difference, the file's existence reliability is determined by comparing it with a probability threshold. If the file's existence reliability meets the threshold, the room affiliation of the file can be immediately determined; if not, the following steps are continued: Flow state fusion: based on statistical calculations of the state flow process, the file's state flow probability is determined. Based on the file's state flow probability, the flow state fusion probability is determined through state transition fusion calculation; Flow state judgment: based on the flow state fusion probability, the room affiliation of the file's flow is determined through probability comparison analysis. The room with the higher file flow state fusion probability is selected as the result room. Since firstly, as... Figure 4As shown, this invention uses Bluetooth gateway positioning technology for positioning. 1. A Bluetooth beacon is fixed on the document folder, emitting a standard Bluetooth signal; 2. A Bluetooth gateway (with different appearances) installed in the environment collects the Bluetooth signal and sends it to a Bluetooth positioning engine located in the cloud. The Bluetooth positioning engine is located on a cloud-based network server and includes specific calculation methods and supporting software tools. Simultaneously, Bluetooth beacons are installed on all documents, and Bluetooth gateways are installed in locations requiring positioning. The gateways locate the Bluetooth beacons (documents), determine their room number, and further determine their dwell time, ultimately achieving intelligent and secure management of documents during circulation. Due to the instability of the Bluetooth signal itself and environmental interference, the result may fluctuate, i.e., the positioning result may jump between adjacent rooms on the same floor or between rooms with the same room number on different floors. The so-called document circulation state means that a document will only exist in a limited number of states throughout the entire circulation process, such as... Figure 4 As shown, the states are: State 1), located in the management office; State 2), located in the secretary's office; State 3), located in the leader's office; State 4), located in the storage room. Specifically, State 1), located in the management office: the document is first logged into the management system; State 2), located in the secretary's office: the document will enter and exit the secretary's office multiple times; State 3), located in the leader's office: generally transferred from the secretary's office to the leader's office for processing, and then back to the secretary's office; State 4), located in the storage room: after the case is closed, the document will be stored in the storage room. There is a certain probability of flow between these four states. To solve the above problems, in this case, through... Optimization methods with file flow status constraints, such as probability reliability judgment, flow status fusion, and flow status judgment, are as follows: Based on the file existence probability of a room determined by the positioning prediction model, the maximum and second largest file existence probabilities of all rooms are selected. By comparing the probability thresholds, a new probability is calculated by fusing the file existence probability and the file flow probability. The magnitude of the fused probability is then compared to determine the room to which the file flow belongs. By fusing the existence probability and the flow probability, errors caused by its own error and external interference are overcome, greatly reducing the misjudgment rate of positioning results and improving the accuracy of item positioning.

[0263] This invention addresses the problem in existing technologies where the misjudgment rate of item positioning results is too high due to the influence of self-positioning errors and internal and external interference factors. It overcomes the errors caused by self-errors and external interference, greatly reduces the misjudgment rate of positioning results, and improves the accuracy of item positioning.

[0264] Any technical solution that utilizes the technical solution of this invention, or any similar technical solution designed by those skilled in the art under the inspiration of the technical solution of this invention, and achieves the above-mentioned technical effects, falls within the protection scope of this invention.

Claims

1. An intelligent document location method based on finite state constraints, characterized in that, include: Probability Confidentiality Assessment: Based on the probability of a file's existence in a room determined by the location prediction model, a probability difference is calculated through probability comparison. This probability difference is then compared with a probability threshold to determine the file's credibility. If the file's credibility meets the threshold, the room affiliation of the file can be immediately determined; otherwise, the following steps are continued: State transition fusion: Based on statistical calculations of the state transition process, the probability of file state transition is determined. Based on the file state transition probability, the probability of state transition fusion is determined through state transition fusion calculation. Flow status judgment: Based on the fusion probability of the flow status, the room to which the file flows is assigned is determined through probability comparison analysis, and the room with the higher fusion probability of the file flow status is selected as the result room; The probability of file existence is obtained by collecting Bluetooth data, which involves setting up a Bluetooth beacon on the file's folder and collecting Bluetooth data through a Bluetooth gateway in the environment. The state transition fusion probability also includes statistical calculation of the state transition process. This statistical calculation includes: based on: for all states, the cumulative number of state transitions exiting from each state within a certain period, and the cumulative number of transitions to a specific state during all these state transitions, the file transition probability is ultimately determined through a state transition probability function; the state transition probability function is: ； Where P is the state transition probability; N is the cumulative number of all processes that have transitioned from a certain state in the past; and N1 is the number of processes that have transitioned from this state to a specific state within a certain period. The state fusion probability further includes state transition fusion calculation, which includes: based on the probability of a file transitioning from the previous state to the first room and the probability of a file transitioning from the previous state to the second room; determining the corresponding state fusion probability of a file transitioning to the first room and the corresponding state fusion probability of a file transitioning to the second room through a state fusion probability function; the state fusion probability function is: ； Wherein, POI is the probability of a file flowing to the first room; POJ is the probability of a file flowing to the second room; M1 is the probability of the file with the maximum value corresponding to the first room; M2 is the probability of the file with the second largest value corresponding to the second room; PI is the probability of the file flowing to the first room and the probability of the file flowing to the second room and the probability of the file flowing to the second room. The flow status determination also includes probability comparison analysis, which includes: based on: The probability of merging the flow status of a file when it is transferred to the first room and the probability of merging the flow status of a file when it is transferred to the second room; The accuracy of file transfer is determined by comparing the transfer status. The flow state comparison function is: ； Where PI is the probability of file transfer to the first room and PJ is the probability of file transfer to the second room.

2. The document location method according to claim 1, characterized in that: The probability confidence assessment also includes probability comparison calculation, which includes: based on: The probability M1 of the maximum value file corresponding to the first room exists, and; The probability M2 of the second largest value file corresponding to the second room exists; The probability difference is determined by a probability comparison function; The probability comparison function is: ； Where E is the probability difference; M1 is the probability of the maximum value file existing in all rooms corresponding to the first room; and M2 is the probability of the second largest value file existing in all rooms corresponding to the second room.

3. The document location method according to claim 1, characterized in that: The probability confidence assessment also includes probability threshold comparison, which includes: Based on the probability difference, a threshold comparison function is used. If the probability difference is greater than the probability comparison threshold, the probability credibility meets the standard; otherwise, the probability credibility does not meet the standard. The threshold comparison function is: ； Where E is the probability difference degree; C0 is the probability comparison threshold.

4. The document location method according to claim 1, characterized in that: The flow state fusion probability also includes a flow state amplification calculation, which includes: based on the flow state fusion probability of a file flowing to the first room and the flow state fusion probability of a file flowing to the second room, amplifying the flow state fusion probability through a flow state amplification function; the flow state amplification function includes: ; Wherein, POI is the probability of a file flowing to the first room state; POJ is the probability of a file flowing to the second room state; PI' is the probability of amplified flow state fusion when a file flows to the first room; and PJ' is the probability of amplified flow state fusion when a file flows to the second room. The amplified flow state fusion probability comparison analysis includes: based on the amplified flow state fusion probability of a file flowing to the first room and the amplified flow state fusion probability of a file flowing to the second room, determining the specific room for file flow through a flow state amplification comparison function; and selecting the room with the higher file flow state fusion probability as the result room. ； Where PI' is the amplified flow state fusion probability of the file being transferred to the first room; PJ' is the amplified flow state fusion probability of the file being transferred to the second room; Before the probability credibility judgment, the document existence probability is obtained, which includes: data feature extraction, localization model training, and document existence probability calculation. Data feature extraction: Based on the collected Bluetooth data, the raw Bluetooth data is converted into corresponding signal feature vectors through information feature extraction; Positioning model training: Based on room location information and the aforementioned signal feature vectors, a high-speed learning machine model is used for training and learning to obtain a positioning prediction model; Based on the input signal feature vector and the corresponding positioning prediction model, the probability of the existence of files in each room is obtained through the calculation method of the high-speed learning machine. The Bluetooth data acquisition also includes: acquiring room calibration data, which includes the corresponding Bluetooth wireless signal information of each reference node and acquisition node, as well as the corresponding room location information.

5. The document location method according to claim 1, characterized in that: The location prediction model is as follows: ； Where H is the hidden layer output matrix; B is the output weight matrix; and T is the probability of the existence of the corresponding file in each room.

6. An intelligent document positioning system based on finite state constraints, characterized in that, include: An intelligent document positioning platform that implements the intelligent document positioning method as described in any one of claims 1 to 5, the intelligent document positioning platform comprising a probability credibility judgment module, a circulation status fusion module, and a circulation status judgment module; The probability credibility judgment module is used to: determine the probability of a file's existence in a room based on the location prediction model; calculate the probability comparison difference through probability comparison; determine the credibility of the file's existence by comparing it with a probability threshold based on the probability comparison difference; and determine whether the credibility of the file's existence meets the standard. If the credibility of the file's existence probability meets the standard, the file's ownership is directly determined; otherwise, the subsequent steps are continued. The flow state fusion module is used to: determine the file state flow probability based on statistical calculations of the state flow process; and determine the flow state fusion probability based on the file state flow probability through state transition fusion calculations. The circulation status judgment module is used to: determine the room affiliation of a file circulation based on the circulation status fusion probability through probability comparison analysis, and select the room with the higher circulation status fusion probability as the result room.

Images