Technical supervision and evaluation system and method based on big data multi-source fusion

Abstract

The invention discloses a technical supervision and evaluation system and method based on big data multi-source fusion, and relates to the technical field of technical supervision and management. Compared with the previous technical supervision means, the method solves the problems of low data quality, complex integration and high potential safety hazard caused by multisource of data in the supervision process in the prior art, and the problems of low working efficiency, untimely problem discovery, lagging processing progress tracking and limited rectification means in the supervision process in the prior art; the multi-source data fusion model based on Kalman filtering is used for fusion of multi-source data in the technical supervision process, the accuracy of the fusion result can be optimized, potential safety hazards are effectively reduced, and the model is high in calculation efficiency and suitable for a real-time application scene; a main equipment technical supervision analysis model is established based on multiple dimensions, quantification and trend analysis are carried out, an analysis result notification mechanism is established, technical supervision key problems are issued regularly, risk prompting is carried out in time, corresponding decision implementation can be made in an auxiliary mode, and the basis efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of technology supervision management technology, and in particular to a technology supervision and evaluation system and method based on big data multi-source fusion. Background Art

[0002] Equipment technical supervision plays an important role in safeguarding public interests by ensuring product quality and safety, maintaining market order, promoting technological innovation and strengthening corporate social responsibility. Existing means of technical supervision include offline on-site inspections and online inquiries and evaluations. In the process of existing technical supervision, collecting multi-source data for supervision can enrich data resources and provide a more comprehensive perspective, but it also brings problems and challenges in terms of data quality, integration complexity, and safety hazards. Effectively managing and utilizing multi-source data is one of the important issues currently facing the field of data management and analysis. In addition, existing technologies have many problems such as low work efficiency, untimely problem discovery, delayed tracking of processing progress, and limited means of rectification. The above problems have an adverse impact on the safe and stable operation of the unit.

[0003] In order to solve the above problems, the present invention proposes a technology supervision and evaluation system and method based on big data multi-source fusion. Summary of the Invention

[0004] The purpose of the present invention is to propose a technology supervision and evaluation system and method based on big data multi-source fusion to solve the problems raised in the background technology:

[0005] The multi-source nature of data in the existing technical supervision process leads to low data quality, complex integration, and high security risks. In addition, the existing technical supervision process is inefficient, problems are not discovered in a timely manner, processing progress tracking is delayed, and rectification measures are limited.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The technical supervision and evaluation system based on multi-source integration of big data includes:

[0008] Data acquisition module: used to collect dynamic data of the entire equipment process based on sensors;

[0009] Data fusion module: used to pre-process and fuse the collected multi-source dynamic data;

[0010] Supervision and analysis module: used to establish supervision and analysis models for key equipment and conduct quantitative and trend analysis;

[0011] Supervision and evaluation module: used to establish a notification mechanism for analysis results, provide corresponding risk warnings, and assist in decision-making;

[0012] The data fusion module includes a data preprocessing unit and a data fusion unit;

[0013] The data preprocessing unit is used to preprocess the collected multi-source dynamic data;

[0014] The data fusion unit is used to perform fusion processing on the data;

[0015] The supervision and evaluation module includes a result notification unit and a decision-making support unit;

[0016] The result notification unit is used to make corresponding result notifications based on the quantitative and trend analysis results of the supervision and analysis module;

[0017] The auxiliary decision-making unit is used to generate a corresponding adjustment strategy according to the notification result.

[0018] The technical supervision and evaluation method based on big data multi-source fusion includes the following steps:

[0019] S1: Collect dynamic data of the entire equipment process based on sensors;

[0020] S2: Preprocess and fuse the collected multi-source dynamic data;

[0021] S3: Establish a key equipment supervision and analysis model to conduct quantitative and trend analysis;

[0022] S4: Establish a mechanism for reporting analysis results, provide corresponding risk warnings, and assist in decision-making.

[0023] Preferably, the fusion process in S2 is based on a multi-source data fusion model of Kalman filtering to fuse multi-source data, specifically as follows:

[0024] Based on the M data sources collected, each data source provides a measurement value about the system state; the system state is represented by the vector X k Represents, where k represents the time step;

[0025] Taking the equipment state as a dynamic system, its state equation and observation equation are expressed as:

[0026] Equation of state:

[0027] X k =AX k-1 +BU k +W k

[0028] Among them, X k is the device state at time k; X k-1 is the device state at time k-1; A is the state transfer matrix; B is the control input matrix; U k is the control input; Wk is the process noise;

[0029] Observation equation:

[0030] Z k m =H k m X k +V k m

[0031] Among them, Z k m is the multi-source data observation value for the mth data source at time k; H k m is the observation matrix; V k m is the observation noise;

[0032] According to the state transition model of the system, use the optimal state estimate x at the previous moment k-1 , predict the current state x k-1 - :

[0033] The Kalman filter assumes that the state of the system evolves linearly, which can be expressed as follows:

[0034] x k-1 - =A k x k-1 +B k U k

[0035] Among them, A k is the state transition matrix, which describes the state change of the system from time k-1 to time k; B k is the control input matrix;

[0036] Predict the covariance matrix P of the current state k - :

[0037] P k - =A k P k-1 A k T +Q k

[0038] Among them, P k-1 is the covariance matrix of the previous moment, Q k is the covariance matrix of the process noise;

[0039] Update the gain matrix K k :

[0040] K k =P k - H k T (H k P k - H k T +R k ) -1

[0041] Among them, H k is the observation matrix; T is the transpose of the vector; R k is the covariance matrix of the measurement noise;

[0042] Combined predicted value x k - and the new measurement z k , update the state estimate x k , the formula is:

[0043]

[0044] Update the covariance matrix P of the state estimate k :

[0045]

[0046] Where I is the identity matrix;

[0047] Through the prediction and update steps of Kalman filtering, the estimated value of the device state is continuously updated to achieve the fusion of multi-source data.

[0048] Preferably, in S3, a main equipment technical supervision analysis model is established based on the equipment type and supervision factor dimensions. The establishment of the model includes the following steps:

[0049] S3.1: Based on the technical parameters and operating characteristics of the equipment, define fuzzy sets to describe the status of the equipment. Define a corresponding membership function for each fuzzy set to determine the degree to which the input data belongs to each fuzzy set.

[0050] S3.2: Use a multi-layer neural network from deep learning to construct an encoder to compress the input device data into a low-dimensional latent representation. Also use a multi-layer neural network to construct a decoder to reconstruct the original input data from the latent representation.

[0051] S3.3: Training autoencoder model based on improved ant colony algorithm;

[0052] S3.4: During training, the output of the fuzzy logic is fused with the intermediate layer representation of the autoencoder model;

[0053] S3.5: Combine autoencoder models and fuzzy logic for device anomaly detection;

[0054] S3.6: As new data is collected, update the model parameters regularly to adapt to changes in the equipment's operating status.

[0055] Preferably, in S3.1, the membership function of each fuzzy set is determined, the boundary values ​​of the variables are defined, the fuzzy rules are defined, and the triggering strength of each rule is calculated; for each output fuzzy set, a weighted average is performed according to the triggering strength to obtain the membership function of the output fuzzy set: and defuzzification processing is performed to obtain the output value of the fuzzy logic.

[0056] Preferably, the autoencoding model is constructed in S3.2 as follows:

[0057] The encoder is constructed as follows:

[0058] Based on the fusion processing input data C, the encoder consists of L layers of neural networks, and the output of the lth layer is:

[0059] h l =σ(ω l h l-1 +b l )

[0060] Among them, ω l is the weight matrix, b l is the bias vector, σ is the activation function; h l-1 is the output of the l-1 layer;

[0061] The output q of the last layer encoder is the potential representation:

[0062] q=h L ;

[0063] Among them, h L is the output of the Lth layer;

[0064] The decoder is constructed as follows:

[0065] The decoder consists of L layers of neural networks, and the input of the lth layer is h l-1 ', output h l 'as follows:

[0066] h l '=σ'(ω l 'h l-1 '+b l ');

[0067] Among them, ω l ' is the weight matrix of the decoder, b l' is the decoder's bias vector, σ' is the decoder's activation function

[0068] The final decoder output is the reconstructed data c = h L '.

[0069] Preferably, in S2 and S3.3, the covariance matrix Q in the multi-source data fusion model of Kalman filtering based on the improved ant colony algorithm is k Parameter values, covariance matrix R of measurement noise k Parameter values, as well as parameters in the autoencoder model are trained; the improved ant colony algorithm is specifically as follows:

[0070] Define the upper and lower limits of the parameters and initialize the pheromone concentration. The pheromone concentration corresponds to the accuracy of the autoencoder model reconstructing the data corresponding to the parameter value at this time.

[0071] At the tth iteration, the jth parameter changes from parameter value u to the next parameter value v, and the accuracy of the autoencoder model reconstructing the data corresponding to the initially released parameter value is set to s uv and inspiration information β uv , and set the transition probability to α is the accuracy heuristic factor, β is the expectation heuristic factor, d uv Indicates the difference from the parameter value u to the next parameter value v, r uv is the expected heuristic function, then:

[0072]

[0073] r uv (t) = γ(d uη +d ηv ) / d uv

[0074] When t≠σ, where σ is the maximum number of iterations:

[0075] γ=(t+σ) / σ

[0076] When t = σ:

[0077] γ=1 / σ

[0078] Among them, τ is the parameter value; allowd is the set of allowed values ​​of the parameter; γ is the adaptive adjustment factor; d uη represents the difference from the parameter value u to the next parameter value η; d ηv represents the difference from the parameter value η to the next parameter value v;

[0079] The accuracy of the autoencoder model reconstructed data corresponding to the parameter values ​​is updated as follows:

[0080] suv (t+1)=(1-μ)s uv (t)+Δs uv (t),s uv ∈[s min ,s max ]

[0081]

[0082] Δs best =1 / f(s best )

[0083] Among them, f(s best ) represents the optimal solution; s max and s min s respectively uv (t) The upper and lower limits.

[0084] Preferably, the S3.4 is as follows:

[0085] The intermediate representation of the g-th layer of the encoder is h g , the output of fuzzy logic is F, and the fusion process is as follows:

[0086] h g '=h g +δF

[0087] Among them, δ is the fusion coefficient.

[0088] Preferably, in S3.5, the reconstruction error of the input data after passing through the autoencoding model is calculated, and an error threshold is set. If the reconstruction error value is greater than the error threshold, it is preliminarily determined that the equipment is abnormal; if the fuzzy logic determines that the equipment status is within the abnormal range, it is finally determined that the equipment is abnormal.

[0089] Preferably, S4 performs several types of notifications based on the analysis results in S3, regularly publishes technical supervision issues, and issues corresponding risk warnings; and also generates corresponding adjustment strategies based on the reported issues for auxiliary decision-making.

[0090] Compared with the existing technology, the present invention provides a technology supervision and evaluation system and method based on big data multi-source fusion, which has the following beneficial effects:

[0091] The present invention uses a multi-source data fusion model based on Kalman filtering to fuse multi-source data in the technical supervision process, which can effectively combine information from different data sources, thereby optimizing the accuracy of the fusion results and effectively reducing safety hazards. The model has high computational efficiency and is suitable for real-time application scenarios. It also uses the fused dynamic data as the analysis object, establishes a technical supervision analysis model for major equipment based on multiple dimensions, conducts quantitative and trend analysis, establishes an analysis result notification mechanism, regularly publishes key technical supervision issues, provides risk warnings in a timely manner, and can assist in making corresponding decisions and implementation, thereby improving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 This is the system block diagram mentioned in Example 1 of the present invention;

[0093] Figure 2 This is a flow chart of the method mentioned in Example 2 of the present invention.

[0094] Meaning of the marks in the figure:

[0095] 100. Data acquisition module; 200. Data fusion module; 210. Data preprocessing unit; 220. Data fusion unit; 300. Supervision and analysis module; 400. Supervision and evaluation module; 410. Result notification unit; 420. Decision support unit. DETAILED DESCRIPTION

[0096] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0097] The present invention uses a multi-source data fusion model based on Kalman filtering to fuse multi-source data in the process of technical supervision, which can effectively combine information from different data sources, thereby optimizing the accuracy of the fusion results and effectively reducing safety hazards. The model has high computational efficiency and is suitable for real-time application scenarios. It also uses the fused dynamic data as the analysis object, establishes a technical supervision analysis model for major equipment based on multiple dimensions, conducts quantitative and trend analysis, establishes an analysis result notification mechanism, regularly publishes key technical supervision issues, provides risk warnings in a timely manner, and can assist in making corresponding decisions and implementation, thereby improving efficiency. Specifically, it includes the following contents.

[0098] Example 1:

[0099] See also Figure 1 The present invention provides a technology supervision and evaluation system based on big data multi-source fusion, including:

[0100] Data acquisition module 100: used to collect dynamic data based on the station-side sensors;

[0101] Data fusion module 200: used for pre-processing and fusion processing of collected multi-source dynamic data;

[0102] Supervision and analysis module 300: used to establish a supervision and analysis model for key equipment and perform quantitative and trend analysis;

[0103] Supervision and evaluation module 400: used to establish a notification mechanism for analysis results, provide corresponding risk warnings, and assist in decision-making;

[0104] The data fusion module 200 includes a data pre-processing unit 210 and a data fusion unit 220;

[0105] The data pre-processing unit 210 is used to pre-process the collected multi-source dynamic data;

[0106] The data fusion unit is used to perform fusion processing on the data;

[0107] The supervision and evaluation module 400 includes a result notification unit 410 and a decision support unit 420;

[0108] The result notification unit 410 is used to make corresponding result notifications based on the quantitative and trend analysis results of the supervisory analysis module 300;

[0109] The auxiliary decision unit 420 is used to generate a corresponding adjustment strategy according to the notification result.

[0110] Example 2:

[0111] Reference Figure 2 The present invention provides a technical supervision and evaluation method based on multi-source fusion of big data, comprising the following steps:

[0112] S1: collects dynamic data based on the station-side sensors; the details are as follows:

[0113] The station-side sensors collect dynamic data based on full-process technical supervision and special supervision issues, technical compliance assessment scores, supplier performance evaluation and bad behavior, operational defects and failures, various online monitoring devices, video equipment, etc.

[0114] S2: Preprocess and fuse the collected multi-source dynamic data; the details are as follows:

[0115] The collected multi-source dynamic data is processed through data cleaning, data normalization and data integration, specifically: processing missing data, erroneous data and redundant data; standardizing the expression and content of fields; integrating data from multiple stages and disciplines into a unified format and standardized data wide table.

[0116] Then, the multi-source data fusion model based on Kalman filtering is used to fuse the multi-source data:

[0117] Based on the M data sources collected, each data source provides a measurement value about the system state; the system state is represented by the vector X k Represents, where k represents the time step;

[0118] Taking the equipment state as a dynamic system, its state equation and observation equation are expressed as:

[0119] Equation of state:

[0120] X k =AX k-1 +BU k +W k

[0121] Among them, X k is the device state at time k; X k-1 is the device state at time k-1; A is the state transfer matrix; B is the control input matrix; U k is the control input; W k is the process noise;

[0122] Observation equation:

[0123] Z k m =H k m X k +V k m

[0124] Among them, Z k m is the multi-source data observation value for the mth data source at time k; H k m is the observation matrix; V k m is the observation noise;

[0125] According to the state transition model of the system, use the optimal state estimate x at the previous moment k-1 , predict the current state x k-1 - :

[0126] The Kalman filter assumes that the state of the system evolves linearly, which can be expressed as follows:

[0127] x k-1 - =A k x k-1 +B k U k

[0128] Among them, A kis the state transition matrix, which describes the state change of the system from time k-1 to time k; B k is the control input matrix;

[0129] Predict the covariance matrix P of the current state k - :

[0130] P k - =A k P k-1 A k T +Q k

[0131] Among them, P k-1 is the covariance matrix of the previous moment, Q k is the covariance matrix of the process noise;

[0132] Update the gain matrix K k :

[0133] K k =P k - H k T (H k P k - H k T +R k ) -1

[0134] Among them, H k is the observation matrix; T is the transpose of the vector; R k is the covariance matrix of the measurement noise;

[0135] Combined predicted value x k - and the new measurement z k , update the state estimate x k , the formula is:

[0136]

[0137] Update the covariance matrix P of the state estimate k :

[0138]

[0139] Where I is the identity matrix;

[0140] Through the prediction and update steps of Kalman filtering, the estimated value of the device state is continuously updated to achieve the fusion of multi-source data.

[0141] The covariance matrix Q of the process noise is obtained based on the ant colony algorithm. k and the covariance matrix R of the measurement noise k .

[0142] The Kalman filter algorithm, which fuses multi-source data, can handle the uncertainty and noise between these data sources, providing more accurate estimates than simple or weighted averages. The Kalman filter algorithm also boasts high computational efficiency, making it suitable for real-time applications. It can quickly track and update system status in real-time even when data sources are changing dynamically.

[0143] The multi-source data fusion model based on Kalman filtering performs multi-source data fusion process in detail as shown in Table 1:

[0144] Table 1 Multi-source data fusion results

[0145] Assessment Project Evaluation Metrics Data Source 1 Data Source 2 Data Source 3 Fusion results Equipment performance Operating efficiency (%) 85 88 90 87.67 Equipment performance Stability (number of failures / month) 2 1 3 2 Equipment reliability Failure frequency (times / year) 5 4 6 5 Equipment reliability Mean time between failures (hours) 5000 5500 4800 5100 Maintenance costs Maintenance cost (10,000 yuan / year) 10 8 12 10 Maintenance costs Spare parts consumption (pieces / month) 15 12 18 15 Security risks Accident rate (%) 1.5 1.2 1.8 1.5 Security risks Number of illegal operations (times / month) 3 2 4 3 Energy saving and environmental protection Energy consumption level (kWh / hour) 120 110 130 120 Energy saving and environmental protection Emission compliance rate (%) 90 92 88 90 Personnel Performance Operation proficiency (score / 100) 80 85 75 80 Personnel Performance Training effect (improvement rate%) 10 12 8 10

[0146] S3: Establish a key equipment supervision and analysis model to conduct quantitative and trend analysis; the details are as follows:

[0147] A technical supervision analysis model for major equipment was established based on equipment type, along with the number of problems discovered, number of rectifications, application of supervision clauses, stages, rectification measures, components, and locations. This model combines fuzzy logic and autoencoder models for analysis. Fuzzy logic can handle fuzzy and uncertain information, enabling the model to better interpret and adapt to complex real-world environments and data. Autoencoder models can learn effective representations of data and capture important features within the data. Combined with fuzzy logic, this allows for more accurate data modeling and prediction, especially when data is unevenly distributed or has fuzzy boundaries. Furthermore, autoencoder models can serve as a feature extraction tool, providing better input features for fuzzy logic. This optimization can improve the performance of fuzzy logic systems, such as increasing classification or decision-making accuracy.

[0148] The establishment of the main equipment technical supervision analysis model includes the following steps:

[0149] S3.1: Based on the equipment's technical parameters and operating characteristics, define fuzzy sets to describe the equipment's status. For example, define fuzzy sets such as "low temperature," "moderate temperature," and "high temperature." Define membership functions for each fuzzy set to determine the degree to which the input data belongs to each fuzzy set. These functions can include linear functions, triangular functions, and other membership functions, and define corresponding boundary values.

[0150] S3.2: Use a multi-layer neural network from deep learning to construct an encoder to compress the input device data into a low-dimensional latent representation. Also use a multi-layer neural network to construct a decoder to reconstruct the original input data from the latent representation.

[0151] S3.3: Training the Autoencoder Model

[0152] Loss function: A loss function such as mean square error (MSE) is used to measure the reconstruction error.

[0153] Optimization algorithm: Use the improved ant colony algorithm to update the model parameters to minimize the loss function.

[0154] S3.4: During training, the output of the fuzzy logic is fused with the intermediate layer representation of the autoencoder model; for example, the degree of fuzziness of the device state determined by the fuzzy logic is added as a regularization term to the loss function of the autoencoder model, or it is fused with the intermediate layer representation of the autoencoder model.

[0155] S3.5: Calculate the reconstruction error of the input data after passing through the autoencoder model and set an error threshold. If the reconstruction error value is greater than the error threshold, it is preliminarily determined that the device is abnormal. If the fuzzy logic determines that the device status is within the abnormal range, it is finally determined that the device is abnormal.

[0156] S3.6: As new data is collected, update the model parameters regularly to adapt to changes in the equipment's operating status.

[0157] For example, suppose we have temperature data from a device. Fuzzy logic defines that the range of "low temperature" is [0,50], and the membership function is linearly decreasing; the range of "moderate temperature" is [40,70], and the membership function is triangular; the range of "high temperature" is [60,100].

[0158] The encoder and decoder of the autoencoder model each consist of a three-layer, fully connected neural network. During training, temperature data is input into the model while also considering the temperature status information output by the fuzzy logic. If the error in the newly input temperature data after reconstruction is large at a certain moment, and the fuzzy logic determines that the temperature is in the "high" range at that moment, this may indicate a device anomaly.

[0159] This autoencoding model that combines fuzzy logic and deep learning can fully utilize the ability of fuzzy logic to handle uncertainty and expert knowledge, as well as the feature learning and anomaly detection capabilities of the deep learning autoencoding model, to improve the accuracy and reliability of technical supervision analysis of major equipment.

[0160] S4: Establish a reporting mechanism for analysis results, provide corresponding risk warnings, and assist in decision-making. The details are as follows:

[0161] To establish a reporting mechanism for analysis results, refer to the following process:

[0162] Identify the responsible party for reporting: Identify the specific department or person responsible for collecting, collating, analyzing and reporting the results.

[0163] Determine the content of the report: describe in detail the specific analysis results that need to be reported, including relevant data, conclusions, problems found, suggestions or measures, etc.

[0164] Set notification period: Specify the time interval for regular notifications, such as daily, weekly, monthly, or a specific time period.

[0165] Choose the communication method: There are many ways to communicate, such as written reports, emails, meetings, bulletin boards, etc., to ensure that the information can be effectively conveyed to the target audience.

[0166] Establish a receiving and feedback mechanism: clearly define the recipients of notifications, establish corresponding feedback channels, and encourage recipients to raise questions, opinions, or suggestions.

[0167] Data analysis and organization: Analyze and organize the collected data according to established methods and standards to extract valuable information and conclusions.

[0168] Result review and verification: Relevant professionals or superiors will review and verify the analysis results to ensure their accuracy and reliability.

[0169] Prepare a communication report: Prepare a report of the analysis results in a clear and concise manner, highlighting the key points and key information.

[0170] Timely notification and release: The report will be sent to relevant personnel or departments through the selected notification method within the specified time.

[0171] Follow-up and supervision: Follow up on the recipient's actions and measures, and urge them to make corresponding improvements, adjustments or decisions based on the notification results.

[0172] Problem Solving and Adjustment: If the recipient raises questions or requires further explanation, the responsible party should provide timely answers and communication, and make necessary adjustments and improvements to the analysis results or notification methods based on the feedback.

[0173] Recording and archiving: Record and archive the notification process, content, feedback and other information for subsequent review and reference.

[0174] Regularly publish key technical supervision issues, provide timely risk warnings, and assist in the implementation of decisions such as the issuance of warning notices, identification of family defects, and technical transformation of equipment.

[0175] Taking equipment type as the dimension, a technical supervision analysis model for major equipment is established according to the dimensions of the number of problems found, the number of rectifications, the application of supervision clauses, stages, rectification measures, components, and locations. Quantification and trend analysis are performed, and corresponding rectification measures are generated based on the supervision results. For details, please refer to Table 2:

[0176] Table 2 Analysis results based on the main equipment technical supervision analysis model

[0177]

[0178] The table shows the problems found during the supervision of the entire operation process of different equipment, and can generate corrective measures in a timely manner to reduce the risk of equipment operation.

[0179] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A technology supervision and evaluation system based on multi-source fusion of big data, characterized by: include: Data acquisition module (100): used for collecting dynamic data of the entire process of the equipment based on sensors; Data fusion module (200): used for pre-processing and fusion processing of collected multi-source dynamic data; Supervision and analysis module (300): used to establish a supervision and analysis model for key equipment and perform quantitative and trend analysis; Supervision and evaluation module (400): used to establish a notification mechanism for analysis results, provide corresponding risk warnings, and assist in decision-making; The data fusion module (200) includes a data preprocessing unit (210) and a data fusion unit (220); The data preprocessing unit (210) is used to preprocess the collected multi-source dynamic data; The data fusion unit is used to perform fusion processing on the data; The supervision and evaluation module (400) includes a result notification unit (410) and a decision-making assistance unit (420); The result notification unit (410) is used to make corresponding result notifications based on the quantitative and trend analysis results of the supervisory analysis module (300); The auxiliary decision unit (420) is used to generate a corresponding adjustment strategy according to the notification result.

2. The technology supervision and evaluation method based on big data multi-source fusion according to claim 1 is characterized in that: The steps include: S1: Collect dynamic data of the entire equipment process based on sensors; S2: Preprocess and fuse the collected multi-source dynamic data; S3: Establish a key equipment supervision and analysis model to conduct quantitative and trend analysis; S4: Establish a mechanism for reporting analysis results, provide corresponding risk warnings, and assist in decision-making.

3. The technology supervision and evaluation method based on big data multi-source fusion according to claim 2 is characterized in that: The fusion process in S2 is based on the multi-source data fusion model of Kalman filtering to fuse multi-source data, specifically as follows: Based on the M data sources collected, each data source provides a measurement value about the system state; the system state is represented by the vector X k Represents, where k represents the time step; Taking the equipment state as a dynamic system, its state equation and observation equation are expressed as: Equation of state: X k =AX k-1 +BU k +W k Among them, X k is the device state at time k; X k-1 is the device state at time k-1; A is the state transfer matrix; B is the control input matrix; U k is the control input; W k is the process noise; Observation equation: in, is the multi-source data observation value for the mth data source at time k; is the observation matrix; is the observation noise; According to the state transition model of the system, use the optimal state estimate x at the previous moment k-1 , predict the current state x k-1 - : The Kalman filter assumes that the state of the system evolves linearly, which can be expressed as follows: x k-1 - =A k x k-1 +B k U k Among them, A k is the state transition matrix, which describes the state change of the system from time k-1 to time k; B k is the control input matrix; Predict the covariance matrix P of the current state k - : P k - =A k P k-1 A k T +Q k Among them, P k-1 is the covariance matrix of the previous moment, Q k is the covariance matrix of the process noise; Update the gain matrix K k : K k =P k - H k T (H k P k - H k T +R k ) -1 Among them, H k is the observation matrix; T is the transpose of the vector; R k is the covariance matrix of the measurement noise; Combined predicted value x k - and the new measurement z k , update the state estimate x k , the formula is: Update the covariance matrix P of the state estimate k : Where I is the identity matrix; Through the prediction and update steps of Kalman filtering, the estimated value of the device state is continuously updated to achieve the fusion of multi-source data.

4. The technology supervision and evaluation method based on big data multi-source fusion according to claim 3 is characterized in that: In S3, a main equipment technical supervision analysis model is established based on the equipment type and supervision factor dimensions. The establishment of the model includes the following steps: S3.1: Based on the technical parameters and operating characteristics of the equipment, define fuzzy sets to describe the status of the equipment. Define a corresponding membership function for each fuzzy set to determine the degree to which the input data belongs to each fuzzy set. S3.2: Use a multi-layer neural network from deep learning to construct an encoder to compress the input device data into a low-dimensional latent representation. Also use a multi-layer neural network to construct a decoder to reconstruct the original input data from the latent representation. S3.3: Training autoencoder model based on improved ant colony algorithm; S3.4: During training, the output of the fuzzy logic is fused with the intermediate layer representation of the autoencoder model; S3.5: Combine autoencoder models and fuzzy logic for device anomaly detection; S3.6: As new data is collected, update the model parameters regularly to adapt to changes in the equipment's operating status.

5. The technology supervision and evaluation method based on big data multi-source fusion according to claim 4 is characterized in that: In S3.1, the membership function of each fuzzy set is determined, the boundary values ​​of the variables are defined, the fuzzy rules are defined, and the triggering strength of each rule is calculated; For each output fuzzy set, weighted average is performed according to the trigger intensity to obtain the membership function of the output fuzzy set: and defuzzification is performed to obtain the output value of the fuzzy logic.

6. The technology supervision and evaluation method based on big data multi-source fusion according to claim 4 is characterized in that: The autoencoder model constructed in S3.2 is as follows: The encoder is constructed as follows: The input data of the fusion processing is C, and the encoder consists of L layers of neural networks. The output of the lth layer is h l for: h l =σ(ω l h l-1 +b l ) Among them, ω l is the weight matrix, b l is the bias vector, σ is the activation function; h l-1 is the output of the l-1 layer; The output q of the last layer encoder is the potential representation: q=h L ; Among them, h L is the output of the Lth layer; The decoder is constructed as follows: The decoder consists of L layers of neural networks, and the input of the lth layer is h l-1 ', output h l 'as follows: h l '=σ'(ω l 'h l-1 '+b l '); Among them, ω l ' is the weight matrix of the decoder, b l ' is the decoder's bias vector, σ' is the decoder's activation function The final decoder output is the reconstructed data c = h L '.

7. The technology supervision and evaluation method based on big data multi-source fusion according to claim 6 is characterized in that: In S2 and S3.3, the covariance matrix Q in the multi-source data fusion model of Kalman filtering based on the improved ant colony algorithm is k Parameter values, covariance matrix R of measurement noise k Parameter values, as well as parameters in the autoencoder model are trained; the improved ant colony algorithm is specifically as follows: Define the upper and lower limits of the parameters and initialize the pheromone concentration. The pheromone concentration corresponds to the accuracy of the autoencoder model reconstructing the data corresponding to the parameter value at this time. At the tth iteration, the jth parameter changes from parameter value u to the next parameter value v, and the accuracy of the autoencoder model reconstructing the data corresponding to the initially released parameter value is set to s uv and inspiration information β uv , and set the transition probability to α is the accuracy heuristic factor, β is the expectation heuristic factor, d uv Indicates the difference from the parameter value u to the next parameter value v, r uv is the expected heuristic function, then: r uv (t)=γ(d uη +d ηv ) / d uv When t≠σ, where σ is the maximum number of iterations: γ=(t+σ) / σ When t = σ: γ=1 / σ Among them, τ is the parameter value; allowd is the set of allowed values ​​of the parameter; γ is the adaptive adjustment factor; d uη represents the difference from the parameter value u to the next parameter value η; d ηv represents the difference from the parameter value η to the next parameter value v; The accuracy of the autoencoder model reconstructed data corresponding to the parameter values ​​is updated as follows: s uv (t+1)=(1-μ)s uv (t)+Δs uv (t),s uv ∈[s min ,s max ] Δs best =1 / f(s best ) Among them, f(s best ) represents the optimal solution; s max and s min s respectively uv (t) The upper and lower limits.

8. The technology supervision and evaluation method based on big data multi-source fusion according to claim 7 is characterized in that: The details of S3.4 are as follows: The intermediate representation of the g-th layer of the encoder is h g , the output of fuzzy logic is F, and the fusion process is as follows: h g ’=h g +δF Among them, δ is the fusion coefficient.

9. The technology supervision and evaluation method based on big data multi-source fusion according to claim 8 is characterized in that: In S3.5, the reconstruction error of the input data after passing through the autoencoder model is calculated, and an error threshold is set. If the reconstruction error value is greater than the error threshold, it is preliminarily determined that the device is abnormal. If the fuzzy logic determines that the device status is within the abnormal range, it is ultimately determined to be a device abnormality.

10. The technology supervision and evaluation method based on big data multi-source fusion according to claim 9 is characterized in that: The S4 makes notifications in several ways based on the analysis results in the S3, regularly publishes technical supervision issues, and provides corresponding risk warnings; it also generates corresponding adjustment strategies based on the notification issues for auxiliary decision-making.

Images