Heterogeneous data regularization method for multi-source information fusion

Abstract

The invention relates to the technical field of data processing, and discloses a heterogeneous data regularization method for multi-source information fusion, which comprises the following steps of: arranging distributed sensors in a plurality of processing stages in a sewage treatment process, and synchronously acquiring detection data of all the distributed sensors through an access protocol and a time sequence; performing semantic analysis on the detection data, performing label normalization on the detection data, and completing semantic label standardization; processing the detection data after semantic standardization to obtain metadata; constructing a data cross-layer representation model based on transfer learning, and performing feature extraction and semantic embedding on metadata to obtain uniform vectorization expression; and according to the vectorization expression, constructing an information knowledge graph oriented to the whole flow of sewage treatment. Unified vectorization expression of time sequence, semantics and structural features among different data sources is achieved, and the fusibility, interpretability and usability of data are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a heterogeneous data regularization method for multi-source information fusion. BACKGROUND

[0002] With the acceleration of urbanization and the improvement of environmental governance requirements, as an important part of urban water cycle, the operation process of sewage treatment system has gradually evolved towards automation, digitization and intelligence. Modern sewage treatment plants usually cover multiple typical process stages, including grid cleaning, sand settling, activated sludge treatment, biological denitrification and phosphorus removal, disinfection and sludge thickening and dewatering, etc. Each treatment stage involves real-time monitoring of various environmental factors, physical and chemical parameters and operating conditions.

[0003] In order to support the construction of smart water affairs, a large number of heterogeneous sensors (such as pH sensors, dissolved oxygen probes, flow meters, liquid level meters, ammonia nitrogen analyzers, etc.) are deployed in the sewage treatment system, and data is aggregated through various automation platforms or distributed collection systems. The data generated by these devices and platforms are highly heterogeneous in structure, semantics, units, time frequency, etc., forming a typical "multi-source heterogeneous data system". However, the data collected by different sensors differ in naming, units, time granularity, communication protocols, etc., lack of unified semantic tags and data formats, resulting in data that cannot be directly interoperated. Therefore, most current sewage treatment monitoring platforms still rely on manual setting of variable meaning and processing logic. It is unable to effectively capture the temporal, spatial and logical multi-level associations in the sewage treatment process. It is unable to convert a large amount of real-time data into domain knowledge that can be reasoned and reused.

[0004] Therefore, it is necessary to design a heterogeneous data regularization method for multi-source information fusion to solve the problems existing in the current technology. SUMMARY

[0005] In view of this, the present application provides a heterogeneous data regularization method for multi-source information fusion, aiming to solve the problems existing in the current technology.

[0006] The present application provides a heterogeneous data regularization method for multi-source information fusion, comprising: Distributed sensors are arranged at several treatment stages in the sewage treatment process, and detection data of all the distributed sensors are collected synchronously through access protocols and time sequences; The detection data are semantically analyzed based on semantic matching, and the detection data are labeled and standardized in semantics; The detection data after semantic standardization are processed to obtain metadata containing data source identification, data type, unit specification and time information; construct a data cross-layer representation model based on transfer learning, and perform feature extraction and semantic embedding on the metadata according to the data cross-layer representation model to obtain a unified vectorized expression; construct an information knowledge graph for the whole process of sewage treatment according to the vectorized expression.

[0007] Further, when distributing distributed sensors in several treatment stages of the sewage treatment process, including: In the grid pool, sand pool, biological reaction pool, secondary sedimentation tank, disinfection tank and sludge treatment stage, sensor nodes with temperature, dissolved oxygen, pH, conductivity, turbidity, flow, liquid level, chemical oxygen demand and ammonia nitrogen detection functions are arranged.

[0008] Further, when collecting detection data of all the distributed sensors through access protocol and time sequence synchronization, including: Various sensors are connected to the same data acquisition network using OPC-UA, MQTT or HTTP protocol, and unified data packaging and format standardization are performed through protocol adaptation middleware; Combine Network Time Protocol or IEEE 1588 Precision Time Protocol to synchronize the clocks of distributed sensors.

[0009] Further, based on semantic matching, the detection data is semantically analyzed, the detection data is labeled and standardized, and the semantic label is standardized, including: The original description information of the detection data is extracted through natural language processing, and the semantic feature extraction includes keyword recognition, part-of-speech analysis, entity extraction and semantic nested analysis; Based on vector semantic similarity calculation, semantic distance measurement and upper and lower semantic relationship matching, the detection data is mapped to the corresponding standard label item, and the same, polysemy and abbreviated data labels are normalized; The normalized label is associated with the corresponding physical meaning, unit system and belonging processing unit, and the structured semantic label item is output.

[0010] Further, the detection data after semantic standardization is processed to obtain metadata containing data source identification, data type, unit specification and time information, including: The semantic standardized detection data is field-decomposed according to the preset structure template, and at least: The data source identification field is used to record the sensor device number, the belonging processing unit and its geographic location code; The data type field is used to indicate the physical property type of the detection value; The unit specification field is used to unify the physical unit corresponding to the numerical value; a time information field for recording original collection time, system receiving time and synchronization timestamp; all fields are packaged as structured metadata entries, represented in JSON, XML or RDF semantic format, and the original data mapping index of the metadata is preserved.

[0011] Further, when constructing the data cross-layer representation model based on transfer learning, the following steps are included: a dual-branch hybrid model architecture composed of a pre-trained semantic encoder and a graph neural network is adopted, the pre-trained semantic encoder is used to process the structured semantic labels contained in the metadata to generate an initial semantic embedding vector; the graph neural network takes multiple wastewater treatment process units and their upstream and downstream relationships as nodes and edges to establish a process-level structure graph, and based on the input metadata, feature propagation and aggregation are performed on the graph to obtain a structured embedding vector.

[0012] Further, when constructing the data cross-layer representation model based on transfer learning, the following steps are included: initial model parameters are obtained by training in a wastewater treatment project that has been annotated with semantics and mapped with process structure; based on a domain adaptation transfer strategy, the model is transferred to a target wastewater treatment scenario and adjusted through label samples; the domain adaptation transfer strategy process includes feature distribution alignment, maximum mean difference regularization and semantic label consistency constraint.

[0013] Further, when the data cross-layer representation model is used to extract features and embed semantics from the metadata to obtain a unified vector representation, the following steps are included: the data source identification field, data type field, unit specification field and time information field in the metadata are independently encoded in a subspace, and word embedding, position embedding and multi-scale time convolution network are used to extract features and model time series for each field; after fusing the embedding vectors of each field, dimension alignment and feature compression are performed to obtain the vector representation, which contains information substructures for representing physical semantics, processing stage position and dynamic change trend.

[0014] Further, when constructing an information knowledge graph for the entire wastewater treatment process based on the vector representation, the following steps are included: entity elements in the vector representation are mapped to nodes in the knowledge graph, including process units, sensor devices, monitoring parameters, time segments and control instructions; directed edge relationships between nodes are established according to the wastewater treatment process logic and process structure, and timestamps or working conditions are added as attributes of the edges; Obtain the information knowledge graph facing the whole process of sewage treatment.

[0015] Further, after the information knowledge graph facing the whole process of sewage treatment is constructed according to the vectorization expression, the method further comprises the following steps: Warning is performed according to the information knowledge graph facing the whole process of sewage treatment. The dynamic state of each node and the associated edge relationship in the graph is updated, and the drift trend of the parameter node is identified. The node state in the current graph is scored in real time according to the historical operation data, a warning mark is triggered when the score result exceeds a threshold value, the abnormal influence propagation path is determined through graph path backtracking, and warning is performed according to the abnormal influence propagation path.

[0016] Compared with the prior art, the beneficial effects of the present application are that by arranging distributed sensors at each processing stage and adopting a unified access protocol and a time sequence synchronization mechanism, the spatiotemporal consistent collection of raw data is realized, the label normalization and standardization are completed through semantic matching, then the metadata features are extracted through structured coding, the data cross-layer representation model based on transfer learning is constructed, the unified vectorization expression of the time sequence, semantic and structural features between different data sources is realized, the information knowledge graph facing the whole process is constructed based on the vector representation, the data fusion, interpretability and usability are improved, the traditional data island pattern is broken, and unified data support is provided for intelligent analysis, abnormal warning, process optimization and other core functions of smart water. BRIEF DESCRIPTION OF DRAWINGS

[0017] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of preferred embodiments and are not meant to limit the present application. Furthermore, the same reference numerals in different drawings identify the same components. In the drawings: Figure 1 A flowchart of the multi-source information fusion heterogeneous data regularization method provided by the embodiment of the present application. DETAILED DESCRIPTION

[0018] Exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0019] In the traditional existing sewage treatment system, the multi-source heterogeneous data system leads to poor data interoperability. Due to the differences in naming rules, communication protocols, unit systems and time synchronization mechanisms of different sensor devices, the data tags lack unified semantic mapping standards, and the multi-dimensional parameters of the process stage cannot be directly associated. The manual setting of variable meaning and processing logic makes it difficult to analyze the temporal correlation, spatial distribution and process level dependency between data, hindering the conversion of real-time monitoring data into domain knowledge.

[0020] For example, a sewage treatment plant has deployed pH sensors, dissolved oxygen probes and ammonia nitrogen analyzers in the biological reaction tank and secondary sedimentation tank. The pH sensor uses OPC-UA protocol to transmit data, the dissolved oxygen probe reports through MQTT protocol, and the ammonia nitrogen analyzer accesses using HTTP protocol. The detection values of different devices are marked as "pH_1", "DO_Value" and "NH3-N_Concentration" in the data tags, with units of "none", "mg / L" and "ppm" respectively. Due to the lack of unified semantic analysis mechanism, the system cannot automatically identify the synonymous relationship between "DO_Value" and "dissolved oxygen", nor can it convert the ppm unit of "NH3-N_Concentration" to the standard mg / L unit. In addition, precise clock synchronization has not been achieved between devices, resulting in a timestamp deviation of up to seconds between different parameters in the same process stage, making it impossible to accurately establish a dynamic correlation model between dissolved oxygen changes and ammonia nitrogen degradation rate.

[0021] If the above problems are not solved, the semantic ambiguity of multi-source heterogeneous data will make it difficult to model the causal relationship between process parameters, and the prediction error of key water quality indicators will increase. Data islands will limit cross-processing unit data fusion analysis, making it impossible to achieve global optimization of control parameters such as aeration rate and sludge return ratio. Process adjustment decisions rely on offline manual experience and cannot respond in real time to fluctuations in influent load, which may lead to instantaneous exceedance of effluent chemical oxygen demand or sludge bulking and other operational risks.

[0022] To this end, refer to Figure 1 The present application proposes a multi-source information fusion heterogeneous data regularization method, comprising: S100: Distribute distributed sensors in several process stages of the sewage treatment process, and collect detection data of all distributed sensors through access protocols and time synchronization.

[0023] S200: Based on semantic matching, perform semantic analysis on the detection data, label and standardize the semantic labels of the detection data.

[0024] S300: Process the detection data after semantic standardization to obtain metadata containing data source identification, data type, unit specification and time information.

[0025] S400: Construct a data cross-layer representation model based on transfer learning, and perform feature extraction and semantic embedding on the metadata according to the data cross-layer representation model to obtain a unified vectorized expression.

[0026] S500: Construct an information knowledge graph for the whole process of sewage treatment according to the vectorized expression.

[0027] Specifically, the distributed sensor refers to a plurality of sensing devices with independent detection function deployed at different stages of the sewage treatment process, which can be realized by temperature, dissolved oxygen, pH, and conductivity sensors, for real-time acquisition of physical and chemical parameters of each process unit, solving the problem of scattered and heterogeneous data sources.

[0028] Semantic matching refers to the analysis of the original description of the detection data through natural language processing technology, which can be realized by keyword recognition, entity extraction, and semantic nested analysis, for eliminating the naming differences of different sensor data labels and solving the interoperability obstacles caused by semantic ambiguity.

[0029] Metadata refers to structured description entries containing data source identification, data type, unit specification, and time information, which can be realized by encapsulation in JSON and XML formats, for unifying the physical properties and space-time benchmarks of multi-source data, solving the problem of inconsistent data format and unit system.

[0030] The data cross-layer representation model of transfer learning refers to a machine learning architecture that can uniformly encode different levels of data features, which can be realized by a dual-branch hybrid model of pre-trained semantic encoder and graph neural network, for extracting cross-process stage correlation features and solving the defect of heterogeneous data difficult to analyze jointly.

[0031] Vectorized expression refers to converting metadata into embedded vectors in a unified mathematical space, which can be realized by word embedding, position embedding, and multi-scale time convolution network, for representing the physical semantics and dynamic trends of data, solving the problem of multi-dimensional information fusion.

[0032] Information knowledge graph refers to a knowledge base that expresses the entity relationship of the whole process of sewage treatment in a graph structure, which can be realized by constructing process unit nodes and directed edge relationships, for revealing the logical association between parameters and solving the problem of real-time data cannot be converted into inferable knowledge.

[0033] The application converts the dispersed heterogeneous sensor data in the sewage treatment process into a unified vector expression through multi-source information fusion technology, and constructs a cross-process level knowledge graph. The scheme eliminates the difference of data labels through semantic matching, combines the transfer learning model to realize cross-layer feature extraction, and finally forms a structured knowledge system that can support intelligent decision-making, effectively solving the problem that traditional methods rely on manual variable setting and cannot capture multi-level associations.

[0034] The working process and principle of the application are as follows: distributed sensors are arranged at multiple treatment stages of the sewage treatment process, and the detection data of these sensors are collected synchronously through access protocols and time sequences. Then the detection data is subjected to semantic analysis and label normalization to complete semantic label standardization. Then the standardized data is processed to generate metadata containing data source identification, data type, unit specification and time information. A data cross-layer representation model based on transfer learning is constructed, and the model is used for feature extraction and semantic embedding of the metadata to obtain a unified vector expression. Finally, an information knowledge graph for the whole process of sewage treatment is constructed according to the vector expression.

[0035] The original data is collected by distributed sensors to ensure the comprehensiveness of the data source. The semantic analysis and standardization steps eliminate the semantic differences between the data, laying a foundation for subsequent processing. The generation of metadata further standardizes the data structure. The construction and application of the data cross-layer representation model realize the unified expression of heterogeneous data, and the finally formed knowledge graph integrates the information of the whole process of sewage treatment.

[0036] As a preferred embodiment, the scheme of the application is implemented as follows: Distributed sensors are arranged at the treatment stages of the sewage treatment plant, such as the grid pool, sand settling pool, biological reaction pool, secondary sedimentation tank and disinfection tank. These sensors include temperature sensors, dissolved oxygen sensors, pH sensors, conductivity sensors, turbidity sensors, flow meters, liquid level meters, etc. Multiple protocols such as OPC-UA, MQTT and HTTP are used to access these sensors, and protocol adaptation middleware is used for data packaging and format standardization. Network time protocol or IEEE 1588 precision time protocol is used for clock synchronization of the sensors to ensure the time sequence consistency of the data.

[0037] The collected detection data is subjected to semantic analysis. First, the semantic features of the data are extracted through natural language processing technology, including keyword recognition, part-of-speech analysis and entity extraction. Then, based on vector semantic similarity calculation and semantic distance measurement, the data is mapped to standard labels. The standardized labels are associated with physical meaning, unit system and belonging processing unit after the normalization processing of synonyms, polysemous words and abbreviations.

[0038] The standardized data is processed to generate metadata. The metadata includes data source identification (such as sensor number, processing unit to which the sensor belongs), data type (such as physical property type), unit specification, and time information (such as collection time, reception time, and synchronization timestamp). The metadata is encapsulated in JSON or XML format, and the mapping index of the original data is retained.

[0039] A data cross-layer representation model based on transfer learning is constructed. The model is composed of a pre-trained semantic encoder and a graph neural network. The semantic encoder processes the structured semantic labels in the metadata to generate an initial semantic embedding vector. The graph neural network takes the process units of the wastewater treatment process as nodes and the process relationship as edges to construct a process-level structure graph. Feature propagation and aggregation are performed on the graph to obtain a structured embedding vector.

[0040] The constructed model is used for feature extraction and semantic embedding of the metadata. Each field of the metadata is independently encoded in a subspace, and word embedding, position embedding, and multi-scale time convolution network are used for feature extraction and time series modeling. The embedding vectors of each field are fused, dimensionally aligned, and compressed to obtain a unified vector representation.

[0041] An information knowledge graph for the whole process of wastewater treatment is constructed based on the vector representation. The entity elements (such as process units, sensor devices, and monitoring parameters) in the vector representation are mapped to nodes in the graph. According to the logic of the wastewater treatment process and the process structure, the directed edge relationship between the nodes is established, and the time stamp or working condition is attached as the attribute of the edge.

[0042] Through the above scheme, the present application realizes effective fusion and regularization of multi-source heterogeneous data. Through unified semantic analysis and standardized processing, the differences in naming, units, and time granularity of different sensor data are eliminated, and the interoperability of the data is improved. The data cross-layer representation model based on transfer learning can capture the time series, spatial, and logical multi-level correlations in the wastewater treatment process, and convert real-time monitoring data into domain knowledge that can be reasoned and reused. The constructed information knowledge graph for the whole process of wastewater treatment provides a basis for intelligent decision-making, which helps to realize global optimization and real-time response to inflow load fluctuations of process parameters, and reduce the risk of instantaneous COD exceedance of effluent or sludge bulking.

[0043] In some of the above schemes of the present application, distributed sensors are arranged at several treatment stages in the wastewater treatment process to collect detection data. However, due to the large differences in process objectives of different treatment stages, if the sensor types and parameter detection functions are not adapted to specific treatment units, it may lead to omission or redundancy of key process parameters, thereby affecting the integrity and accuracy of subsequent data regularization processing.

[0044] The application further proposes that sensor nodes with temperature, dissolved oxygen, pH, conductivity, turbidity, flow, liquid level, chemical oxygen demand and ammonia nitrogen detection functions are respectively arranged in the grid pool, grit chamber, biological reaction pool, secondary sedimentation tank, disinfection tank and sludge treatment stage.

[0045] Among them, the sensor nodes arranged in the grid pool contain flow and liquid level detection functions, which are used to monitor the water inflow and the accumulation state of intercepted impurities. The sensor nodes of the grit chamber are configured with turbidity and conductivity detection functions to quantify the sand settling effect and ion concentration of the water body. The sensor nodes of the biological reaction pool are integrated with dissolved oxygen, chemical oxygen demand and ammonia nitrogen detection functions to evaluate the oxidation environment and pollutant removal efficiency of the biological degradation process. The sensor nodes of the secondary sedimentation tank are deployed with turbidity and liquid level detection functions to monitor the sludge sedimentation effect and the clarity of the effluent. The sensor nodes of the disinfection tank are equipped with pH and conductivity detection functions to control the dosage of disinfectant and water quality stability. The sensor nodes of the sludge treatment stage contain flow and liquid level detection functions to optimize the control of the sludge dewatering and transportation process.

[0046] Specifically, the functional configuration of the sensor nodes in each treatment unit realizes targeted data acquisition based on its process objectives. For example, the biological reaction pool monitors the oxygen content required for microbial activity in real time through the dissolved oxygen sensor, and quantifies the denitrification efficiency through the ammonia nitrogen sensor, to ensure that the metabolic environment of the biological treatment stage is within the optimal range. The disinfection tank detects the pH value of the water body through the pH sensor, and evaluates the residual concentration of the disinfectant through the conductivity sensor, to avoid secondary pollution caused by excessive dosage. By deploying sensor nodes with functionally adapted functions in each treatment stage, key process parameters of the entire wastewater treatment process can be completely covered, avoiding data regularization bias caused by parameter omission or redundancy. Therefore, the subsequent semantic analysis, metadata generation and knowledge graph construction process can be based on comprehensive and accurate detection data, thereby improving the reliability of the state representation and abnormal early warning of the entire wastewater treatment process.

[0047] As a preferred embodiment, the scheme of the application is implemented as follows: distributed sensors are arranged in each treatment stage of the wastewater treatment plant. Specifically, temperature sensors and pH sensors are installed in the grid pool. Dissolved oxygen sensors and turbidity sensors are installed in the grit chamber. Conductivity sensors, chemical oxygen demand sensors and ammonia nitrogen sensors are installed in the biological reaction pool. Flow sensors are installed in the secondary sedimentation tank. Liquid level sensors are installed in the disinfection tank. Temperature sensors and pH sensors are installed in the sludge treatment stage. These sensor nodes respectively have corresponding detection functions and can monitor the key parameters of each treatment stage in real time.

[0048] By the technical scheme, the application achieves comprehensive monitoring of the whole sewage treatment process. Thus, the operation state of each treatment unit can be grasped in time, and data support is provided for process control and abnormality early warning. Further, the cooperative arrangement of multiple sensors helps to capture the parameter correlation between different treatment stages, and provides a basis for overall optimization of the sewage treatment process.

[0049] In some schemes of the application, the detection data of all distributed sensors is collected synchronously through an access protocol and timing, however, the use of heterogeneous communication protocols by different sensors leads to the inability to uniformly encapsulate the data format, and the lack of a unified clock synchronization mechanism, resulting in chaotic data collection timing.

[0050] The application further proposes that when the detection data of all distributed sensors is collected synchronously through an access protocol and timing, it includes: using OPC-UA, MQTT or HTTP protocol to access the same data collection network for sensors of various types with heterogeneous communication protocols, and performing unified data encapsulation and format standardization through protocol adaptation middleware. The network time protocol or IEEE 1588 precision time protocol is combined to perform unified clock synchronization for distributed sensors.

[0051] Among them, the protocol adaptation middleware has multiple communication protocol conversion modules built-in, which can convert data transmitted by different protocols into data packets of a unified format. For example, when the sensor uses Modbus protocol for transmission, the middleware converts it into JSON format and adds a timestamp field. The clock synchronization module broadcasts synchronization signals to all sensors through a time server, so that the clock error of each node is controlled within the millisecond level. For high-precision scenarios, the IEEE 1588 protocol supporting sub-microsecond synchronization is used to achieve accurate time synchronization through hardware timestamps.

[0052] Specifically, when the dissolved oxygen sensor deployed in the biological reaction tank transmits detection data through the MQTT protocol, the protocol adaptation middleware parses the message payload into structured data, encapsulates it into a standardized data frame after supplementing the device number and process unit code. At the same time, the clock synchronization module sends a time correction instruction to the sensor network every 5 minutes to ensure that the time reference of each node is consistent. For example, the turbidity sensor deployed in a sewage treatment plant transmits data using the HTTP protocol, the middleware converts it into an XML format data packet containing device ID, value, unit, and timestamp, and calibrates the built-in clock of the sensor through the NTP server, so that the time deviation of the 300 sensors in the whole plant is less than 50 milliseconds. This double standardization mechanism aligns the detection data from different manufacturers and different protocols in terms of format and time dimension, providing input data with consistent structure and accurate timing for subsequent semantic analysis and feature extraction.

[0053] As a preferred embodiment, the scheme of the application is implemented as follows: The temperature sensor, dissolved oxygen sensor, and pH sensor are connected to the data acquisition network using the OPC-UA protocol. Specifically, the temperature sensor uses the PT100 model, with a measurement range of 0-100°C and an accuracy of ±0.1°C. The dissolved oxygen sensor uses the fluorescence method, with a measurement range of 0-20mg / L and an accuracy of ±0.1mg / L. The pH sensor uses the glass electrode method, with a measurement range of 0-14 and an accuracy of ±0.01pH. These sensors are connected to the OPC-UA server through the RS485 interface.

[0054] The flow meter and liquid level meter are connected to the data acquisition network using the MQTT protocol. Specifically, the flow meter uses the electromagnetic method, with a measurement range of 0-1000m³ / h and an accuracy of ±0.5%. The liquid level meter uses the ultrasonic method, with a measurement range of 0-10m and an accuracy of ±1mm. These sensors output 4-20mA analog signals, which are converted to MQTT messages by a signal converter and published to the MQTT broker server.

[0055] The chemical oxygen demand analyzer and ammonia nitrogen analyzer are connected to the data acquisition network using the HTTP protocol. Specifically, the chemical oxygen demand analyzer uses the potassium dichromate method, with a measurement range of 0-1000mg / L and an accuracy of ±3%. The ammonia nitrogen analyzer uses the Nessler reagent colorimetric method, with a measurement range of 0-50mg / L and an accuracy of ±3%. These analyzers upload data to the Web server using the HTTP POST method through the Ethernet interface.

[0056] Further, a protocol adaptation middleware is used for unified data packaging and format standardization. Specifically, the middleware is built on the Node-RED open source platform, and JavaScript functions are written to convert data of different protocols into a unified JSON format, including device ID, measurement parameter, value, unit, timestamp, etc.

[0057] Thus, the network time protocol is used for unified clock synchronization of distributed sensors. Specifically, an NTP server is deployed in the data acquisition network, and all sensor devices and servers synchronize time with the NTP server through NTP clients, with a synchronization accuracy of milliseconds.

[0058] Through the above technical solutions, the application realizes unified access and format standardization of heterogeneous sensor data, improves the efficiency and reliability of data acquisition. The unified data format facilitates subsequent processing and analysis, reducing the workload of data conversion and cleaning. Clock synchronization ensures the time consistency of data from different sources, laying a foundation for multi-source data fusion and time series analysis. In addition, the use of open standard communication protocols improves the interoperability and scalability of the system.

[0059] In some schemes of the present application, when the detection data is semantically parsed based on semantic matching, the original data labels have synonymy, polysemy and abbreviation differences, which leads to the fact that the semantic descriptions of different sensor data cannot be uniformly mapped to the standard label system, affecting the accuracy of subsequent metadata generation and knowledge graph construction.

[0060] The present application further proposes that when the detection data is semantically parsed based on semantic matching, the original description information of the detection data is subjected to semantic feature extraction through natural language processing, which includes keyword recognition, part-of-speech analysis, entity extraction and semantic nested parsing. Based on vector semantic similarity calculation, semantic distance measurement and upper and lower semantic relationship matching, the detection data is mapped to the corresponding standard label item, and the synonymy, polysemy and abbreviation data labels are normalized. The normalized labels are associated with the corresponding physical meaning, unit system and belonging processing unit, and the structured semantic label item is output.

[0061] Among them, the semantic feature extraction adopts the word vector model in natural language processing to perform word segmentation and vectorization on the original description text, for example, keyword recognition and entity extraction are realized through Word2Vec or BERT model. The vector semantic similarity calculation measures the semantic correlation degree between different labels through cosine similarity or Euclidean distance, for example, when the vector similarity threshold of "DO" and "dissolved oxygen" is set to 0.85 or more, it is determined as synonymous labels. The upper and lower semantic relationship matching establishes a hierarchical relationship based on the ontology library in the field of sewage treatment, for example, "COD" is mapped to the upper concept "water quality index". The structured semantic label item is stored in the form of triple, for example, <sensor number, standard label name, belonging processing unit>.

[0062] Specifically, the original description information of the detection data is first input into the pre-trained natural language processing model, and the context features are extracted through the bidirectional long short-term memory network to identify key entities such as "pH" and "liquid level". Subsequently, the similarity between the original label vector and the candidate labels in the standard label library is calculated, and when the semantic distance between "TOC" and "total organic carbon" is lower than the preset threshold, they are merged into the same standard label. Further, the upper and lower relationships of the labels are verified through the domain knowledge graph, for example, it is confirmed that "NH3-N" belongs to the "nitrogen compound" category. The finally generated semantic label item contains physical unit conversion rules, for example, "mg / L" is unified to "g / m³", and is associated with the geographic coding of the biological reaction tank. This process realizes the semantic consistency of cross-sensor data by eliminating label ambiguity and redundancy.

[0063] As a preferred embodiment, the schemes of the present application are implemented as follows: The original description information of the detection data is subjected to semantic feature extraction through natural language processing. The semantic feature extraction includes keyword identification, part-of-speech analysis, entity extraction, and semantic nested analysis. For example, for the original description of "pH value 7.2", the keyword identification extracts "pH" and "7.2", the part-of-speech analysis identifies "pH" as a noun and "7.2" as a numerical value, the entity extraction identifies "pH" as a water quality parameter and "7.2" as a specific numerical value, and the semantic nested analysis establishes the attribute-value relationship between "pH" and "7.2".

[0064] Further, based on vector semantic similarity calculation, semantic distance measurement, and upper and lower semantic relationship matching, the detection data is mapped to the corresponding standard label item, and the same, polysemy, and abbreviated data labels are normalized. Specifically, the cosine similarity between the original description and the standard label is calculated using a pre-trained word vector model, and the standard label with the highest similarity is selected. Meanwhile, the upper and lower relationships in the semantic tree are considered, such as "dissolved oxygen" being a lower concept of "water quality parameter". For the abbreviated label "DO", it is mapped to the standard "dissolved oxygen" label through a synonym table.

[0065] In this way, the normalized label is associated with the corresponding physical meaning, unit system, and processing unit, and a structured semantic label item is output. For example, "pH 7.2" is standardized as a structured item of { "parameter": "pH value", "value": 7.2, "unit": "dimensionless", "processing unit": "biological reaction tank"}.

[0066] Through the above technical solutions, the application realizes semantic unification and standardized expression of heterogeneous data. The semantic features of the detection data are accurately extracted, and the ambiguity caused by non-standard naming is eliminated. The standardized semantic labels facilitate subsequent data analysis and knowledge reasoning, and improve the interoperability of the data. The structured semantic label item contains complete context information, laying a foundation for information fusion in the whole process of wastewater treatment.

[0067] In some of the above schemes of the application, although the detection data after semantic matching and standardization achieves label normalization, it lacks a unified structured metadata organization method, making it difficult for subsequent cross-layer representation models to effectively integrate device attributes, physical units, and time dimension information in multi-source heterogeneous data, affecting the accurate mapping of entity relationships in the knowledge graph construction process.

[0068] The application further proposes that the detection data after semantic standardization is field-decomposed according to a preset structure template, at least including a data source identification field, a data type field, a unit specification field, and a time information field, and all fields are encapsulated as structured metadata items, represented in JSON, XML, or RDF semantic format, and the original data mapping index of the metadata is preserved.

[0069] The data source identification field records the data source in the form of a combination of device number, processing unit name and geographic location code, for example, "BIO-01-12.34N / 45.67E" represents the first sensor of the bioreactor tank and its latitude and longitude coordinates. The data type field is matched and labeled by a pre-defined physical property type library, for example, "dissolved oxygen concentration" corresponds to "DO" type code. The unit specification field uses the International System of Units for forced conversion, for example, "mg / L" is uniformly converted to "kg / m³". The time information field records the collection, reception and synchronization time in the format of Coordinated Universal Time, for example, "2023-08-01T08:00:00Z". When the structured metadata entry is in JSON format, a nested key-value pair structure can be constructed, including "source_id", "data_type", "unit", "timestamp" and other primary keys, and each primary key is hung with a subfield. The original data mapping index generates a unique identifier through a hash algorithm, for example, the SHA-256 algorithm is used to calculate the digest of the original detection value.

[0070] Specifically, when the semantic standardized detection data enters the metadata encapsulation stage, the preset structure template first splits the data stream into independent fields. The data source identification field generates a unique identifier containing spatial location information by analyzing the sensor device number and process unit topological relationship, ensuring data traceability. The data type field converts the semantic label to a standardized code by querying the pre-defined physical property mapping table, eliminating naming ambiguity. The unit specification field realizes the normalization processing of numerical values through the unit conversion function library, for example, the flow units of different dimensions are unified to "m³ / s". The time information field aligns the time reference of multi-source data through the time synchronization service and records the complete timestamp sequence. During encapsulation, JSON format uses hierarchical nested structure to organize each field, for example, three sub-keys "collect", "receive" and "sync" are set under the "timestamp" key to store different time nodes. RDF format expresses entity relationship in the form of triple, for example, <sensor001> <hasunit><kg / m³>”. The original data mapping index establishes a bidirectional association between the metadata entry and the original detection data through a hash value, thereby guaranteeing data traceability. Through the above structured processing, the metadata entry can completely carry multi-dimensional information, thereby providing a unified data interface for subsequent feature extraction and knowledge graph construction.

[0071] As a preferred embodiment, the scheme of the present application is implemented as follows: When the detection data after semantic standardization is processed to obtain metadata containing data source identification, data type, unit specification, and time information, the method comprises: The detection data after semantic standardization is field-decomposed according to a preset structure template, and at least the following is extracted and collected: The data source identification field is used to record the sensor device number, the processing unit to which the sensor device belongs, and the geographic position code.

[0072] The data type field is used to indicate the physical property type of the detection value.

[0073] The unit specification field is used to unify the physical unit corresponding to the numerical value.

[0074] The time information field is used to record the original collection time, the system receiving time, and the synchronization time stamp.

[0075] All fields are encapsulated as structured metadata entries, represented in JSON, XML, or RDF semantic format, and the original data mapping index of the metadata is retained.

[0076] Specifically, the following steps can be used to achieve this: First, a preset structured template is defined, which includes four main fields of data source identification, data type, unit specification, and time information.

[0077] Second, each piece of detection data after semantic standardization is parsed to extract relevant information and fill it into the corresponding field: The data source identification field can record the device number in the form of "DO_sensor_01_bio_reactor_A", the processing unit name "biological_treatment", and the geographic coordinates "30.5425°N, 114.3572°E".

[0078] The data type field can be labeled as "dissolved_oxygen" and other physical quantity types.

[0079] The unit specification field is unified as the International System of Units, such as the dissolved oxygen concentration unit specification "mg / L".

[0080] The time information field records the original collection time in the format of "2023-05-01 10:30:15.000", the system receiving time, and the Unix timestamp in the format of "1682912215000".

[0081] Then, the filled structured data is converted to JSON format, for example: { "data_source": { "device_id": "DO_sensor_01_bio_reactor_A", "process_unit": "biological_treatment", "location": "30.5425°N, 114.3572°E" }, "data_type": "dissolved_oxygen", "unit": "mg / L", "time_info": { "collection_time": "2023-05-01 10:30:15.000", "receive_time": "2023-05-01 10:30:15.500", "timestamp": 1682912215000 }, "value": 6.5, "raw_data_index": 1024 } Finally, the JSON format metadata is stored in the database or file system, and the mapping relationship with the original data is established.

[0082] Through the above technical solutions, the application realizes the unified structured expression of heterogeneous data. Thus, the interoperability and parsability of data can be improved, laying a foundation for subsequent data analysis and knowledge graph construction. Further, by retaining the original data mapping, the traceability of the data processing process is ensured. In addition, the use of standard JSON format is conducive to cross-platform data exchange and system integration.

[0083] In some schemes of the application, when constructing the data cross-layer representation model, there is a problem that structured semantic labels and process structure information are difficult to be effectively fused. The traditional single model architecture cannot simultaneously process the semantic features in the metadata and the topological relationship between the process units, resulting in a lack of hierarchy and relevance in feature expression, which affects the entity mapping accuracy of the subsequent knowledge graph.

[0084] The application further proposes to adopt a dual-branch hybrid model architecture composed of a pre-trained semantic encoder and a graph neural network. The pre-trained semantic encoder is used to process the structured semantic labels contained in the metadata to generate initial semantic embedding vectors. The graph neural network takes a plurality of sewage treatment process units and the upstream and downstream relationships therebetween as nodes and edges to establish a process-level structure graph, and performs feature propagation and aggregation on the graph based on the input metadata to obtain structured embedding vectors.

[0085] Among them, the pre-trained semantic encoder adopts a BERT model based on a Transformer architecture, which realizes deep analysis of structured information such as entity types and unit specifications in semantic labels by pre-training on a large amount of industrial text corpus. The graph neural network adopts a graph convolution network or a graph attention network, modeling the biological reaction tank, the secondary sedimentation tank and other process units as graph nodes, and the material flow and process dependency relationship therebetween as edge connections. The dual-branch architecture realizes the fusion of semantic embedding vectors and graph structure features through a cross-modal attention mechanism, for example, a gating fusion module is used to dynamically adjust the weight proportion of the two types of features.

[0086] Specifically, in the model training phase, the pre-trained semantic encoder first vectorizes the semantic labels in the metadata to generate initial semantic embeddings with a dimension of 768. At the same time, the graph neural network constructs an adjacency matrix according to the connection relationship between the process units, and aggregates the feature information of adjacent nodes through three layers of graph convolution operations. In the feature fusion stage, the semantic embedding vector and the graph structure embedding vector are input into a multilayer perceptron for dimension alignment, and after normalization, they are spliced into a joint vector with a dimension of 1024. The joint vector is connected through a residual connection and a layer normalization operation, and finally outputs a structured embedding containing semantic features and topological relationships. For example, in the embedding vector of the sludge treatment unit, it contains not only the semantic feature of "ammonia nitrogen concentration", but also the process association information of this unit with the biological reaction tank and the disinfection tank. This dual-branch architecture effectively solves the limitations of traditional single-modal models in cross-layer feature expression, and the entity mapping accuracy of the subsequent knowledge graph is improved to 92.3%.

[0087] As a preferred embodiment, the schemes of the application are implemented as follows: When constructing the data cross-layer representation model based on transfer learning, a double-branch hybrid model architecture composed of a pre-trained semantic encoder and a graph neural network is adopted. The pre-trained semantic encoder adopts the BERT model to process the structured semantic labels contained in the metadata and generate initial semantic embedding vectors. The graph neural network adopts the graph attention network (GAT) to take multiple sewage treatment process units and the upstream and downstream relationships therebetween as nodes and edges to establish a process-level structure graph. Specifically, the grid pool, grit chamber, biological reaction pool, secondary sedimentation tank, disinfection tank and sludge treatment unit are taken as nodes in the graph, and the material transmission relationship in the process flow is taken as a directed edge. Based on the input metadata, feature propagation and aggregation are performed on the graph to obtain a structured embedding vector.

[0088] Through the above technical solutions, the application realizes unified representation of multi-source heterogeneous data in the sewage treatment process. Thus, data of different sources and formats can be mapped to the same semantic space, facilitating subsequent data analysis and knowledge reasoning. Further, by introducing the graph neural network, the present application can effectively capture the correlation between the process units of the sewage treatment, improving the accuracy and completeness of data representation.

[0089] In some of the above schemes of the application, when constructing the data cross-layer representation model based on transfer learning, direct application of initial model parameters in the target sewage treatment scene may lead to mismatch between semantic labels and process structure mapping relationship, limited model generalization ability, and inability to effectively adapt to the heterogeneous data distribution differences of different sewage treatment projects.

[0090] The application further proposes to obtain initial model parameters by performing source domain training in a sewage treatment project that has been built with semantic annotation and process structure mapping. Based on the domain adaptation transfer strategy, the model is transferred to the target sewage treatment scene and adjusted through label samples. The domain adaptation transfer strategy process includes feature distribution alignment, maximum mean difference regularization and semantic label consistency constraint.

[0091] Among them, the source domain training adopts a sewage treatment project dataset with complete semantic annotation and process structure mapping relationship, and optimizes the model parameters through supervised learning to enable the model to capture the correlation between data features and process logic. In the domain adaptation transfer strategy, the feature distribution alignment aligns the feature spaces of the source domain and the target domain through adversarial training or covariance matching, the maximum mean difference regularization calculates the distribution difference of the data of the two domains using a kernel function and uses it as a regularization term to constrain model optimization, and the semantic label consistency constraint ensures that the semantic labels of the target domain samples remain coherent with the source domain through contrastive learning or label propagation algorithm.

[0092] Specifically, the source domain training stage adopts the labeled data of historical sewage treatment projects, and optimizes the parameters of the pre-trained semantic encoder and graph neural network through the cross-entropy loss function, so that the model can accurately extract the semantic features of the metadata and establish the correlation between the process units. When migrating to the target scene, the feature distribution alignment module is used to map the metadata features of the source domain and the target domain in space, and the maximum mean difference regularization term is used to minimize the difference between the feature distributions of the two domains, thereby reducing the influence of domain bias on the performance of the model. At the same time, the semantic label consistency constraint module compares the similarity of the embedding vectors of the same semantic labels in the source domain and the target domain to ensure the continuity of the label semantics during the migration process. For example, when migrating the dissolved oxygen monitoring parameters of the sludge treatment unit, the feature distribution alignment module matches the distribution of the target domain sensor data with the source domain covariance, the maximum mean difference regularization uses a Gaussian kernel function to calculate the distribution distance of the data in the hidden space of the two domains and adds it to the loss function, and the semantic label consistency constraint module calculates the cosine similarity of the dissolved oxygen label in the target domain and the corresponding label in the source domain to force the model to maintain the consistency of the label semantics. Therefore, the model only needs a small amount of labeled samples to complete the parameter fine-tuning in the target scene, realizes efficient migration across sewage treatment projects, and improves the accuracy and stability of data representation.

[0093] As a preferred embodiment, the scheme of the application is implemented as follows: When constructing the data cross-layer representation model based on transfer learning, the following steps are further included: The initial model parameters are obtained by training the source domain in a sewage treatment project that has been semantically annotated and process structure mapped. Specifically, a stable and well-annotated sewage treatment plant is selected as the source domain, and the initial model is trained using the historical operation data and expert annotation information of the plant. During the training process, the batch stochastic gradient descent algorithm is used to optimize the model parameters, and the learning rate is set to 0.001 with 100 training rounds.

[0094] Based on the domain adaptation transfer strategy, the model is transferred to the target sewage treatment scene and adjusted through label samples. The domain adaptation transfer strategy process includes feature distribution alignment, maximum mean difference regularization, and semantic label consistency constraint. Among them, the feature distribution alignment uses the maximum mean difference (MMD) loss function to calculate the distance between the feature distributions of the source domain and the target domain, and minimizes the distance. The maximum mean difference regularization controls the difference between the feature distributions of the source domain and the target domain by adding the MMD term to the objective function. The semantic label consistency constraint ensures that the model produces consistent prediction results for the same semantic data in the source domain and the target domain by introducing a label prediction consistency loss.

[0095] Further, during the migration process, a progressive migration strategy is adopted. First, the parameters of the pre-trained semantic encoder are frozen, and only the graph neural network part is fine-tuned. Then, the high-level parameters of the semantic encoder are gradually unfrozen for fine-tuning. Finally, the entire model is fine-tuned end-to-end. During the fine-tuning process, the learning rate is set to 0.0001, and the iteration round is 50 rounds.

[0096] Thus, by the above technical solutions, the application realizes cross-scene migration and self-adaptation of sewage treatment data. The model can quickly adapt to new sewage treatment scenarios, reducing the dependence on a large amount of labeled data. At the same time, through feature distribution alignment and semantic consistency constraint, the generalization ability and prediction accuracy of the model in the target domain are improved. In addition, the progressive migration strategy effectively prevents catastrophic forgetting of the model during the migration process, ensuring the stability of the model performance.

[0097] In some of the above schemes of the application, when the data cross-layer representation model extracts features and embeds semantics from the metadata, the encoding methods of different fields fail to effectively separate the information of physical semantics, processing stage position and dynamic change trend, resulting in information mixing and dimension redundancy in vector expression.

[0098] The application further proposes to independently code the data source identification field, data type field, unit specification field and time information field in the metadata in a subspace, and use word embedding, position embedding and multi-scale time convolution network to extract features and model time series from each field. After fusing the embedding vectors of each field, dimension alignment and feature compression are performed to obtain vector expression, which contains information substructures for representing physical semantics, processing stage position and dynamic change trend.

[0099] Among them, independent subspace coding is realized by assigning independent coding channels to each field, for example, the data source identification field uses hash word embedding based on device number and geographic coding, the data type field uses pre-trained word vector mapping based on physical attribute label, the unit specification field uses joint coding of unit conversion coefficient and dimension symbol, and the time information field uses position embedding of segmented timestamp. The multi-scale time convolution network is composed of multiple time series convolution layers with different convolution kernel widths, for example, setting 1 minute, 5 minutes and 15 minutes as three time windows to capture dynamic change patterns of different time granularities. Dimension alignment and feature compression map the embedding vectors of each field to a unified dimension space through a fully connected layer, and then perform dimension reduction through principal component analysis or an autoencoder.

[0100] Specifically, the hash word embedding of the data source identification field converts the device number and the geographic code into a fixed-length vector, avoiding coding conflicts of different devices. The pre-trained word vector of the data type field is obtained by training a corpus in the field of sewage treatment, ensuring that the semantic similarity of the physical property label is maintained in the vector space. The joint encoding of the unit specification field separately normalizes the unit conversion coefficient and the dimension symbol, eliminating the influence of unit system differences on the value. The segmented position embedding of the time information field divides the original collection time, system reception time, and synchronization timestamp into three time segments of hours, minutes, and seconds, and performs sinusoidal position coding respectively. The multi-scale time convolution network adopts a parallel convolution structure, and the outputs of each convolution layer are spliced into a time sequence feature vector after maximum pooling. In the dimension alignment process, the weight matrix of the fully connected layer is dynamically adjusted according to the embedding dimensions of each field to ensure that the vectors of different fields are comparable in a unified space. Feature compression is achieved through the encoder part of the autoencoder, retaining 95% of the variance contribution rate to eliminate redundant dimensions. In the resulting vector representation, physical semantic information is dominated by the embedding vectors of the data type and unit fields, processing stage position information is represented by the embedding vectors of the data source identification field, and dynamic change trend information is described by the time field's time convolution features.

[0101] As a preferred embodiment, the scheme of the application is implemented as follows: the data source identification field in the metadata is subjected to word embedding processing, the device number is converted into a 64-dimensional distributed vector representation through a pre-trained word vector model, and the geographic position code is generated into a 128-dimensional vector based on a polar coordinate system position embedding algorithm. The data type field is mapped through a physical property ontology library and a one-hot encoding is used to generate a category vector. The unit specification field is normalized through an international unit conversion table and a linear projection layer is used to generate a 32-dimensional embedding vector. The time information field is input into a multi-scale time convolution network, which includes three parallel dilated convolution layers with expansion factors of 1, 3, and 5, respectively, for capturing time sequence features at the hour, day, and week levels. After layer normalization, the encoding results of each subspace are fused through a gated attention mechanism, and finally a fully connected layer is used to compress the fused 1024-dimensional vector into a 256-dimensional vector representation.

[0102] Through the above technical scheme, the application realizes unified representation of multi-modal metadata features, solving the semantic fragmentation problem caused by coding method differences in different attribute fields in traditional methods. The independent subspace encoding mechanism effectively isolates the interference of heterogeneous features such as device identification and spatio-temporal information, and accurately captures the dynamic evolution law of sewage treatment parameters through multi-scale time sequence modeling. The feature compression process reduces the redundancy of high-dimensional vectors, enabling the subsequent knowledge graph construction process to efficiently identify the association between entities.

[0103] In some schemes of the above-mentioned schemes of the present application, in the feature extraction and semantic embedding process of the data cross-layer representation model, the heterogeneity of different fields leads to difficulty in feature fusion, and it is difficult to effectively integrate physical semantics, processing stage position and dynamic change trend information, affecting the integrity and consistency of vectorization expression.

[0104] The present application further proposes to independently code the data source identification field, the data type field, the unit specification field and the time information field in the metadata, and use word embedding, position embedding and multi-scale time convolution network to extract features and model time series of each field. After fusing the embedding vectors of each field, dimension alignment and feature compression are performed to obtain vectorization expression, which contains information substructure for representing physical semantics, processing stage position and dynamic change trend.

[0105] Among them, the data source identification field generates a word embedding vector through device number and geographic location coding, mapping the physical location of the device and the association relationship of the process unit. The data type field constructs a semantic vector based on the physical property type label to capture the semantic similarity between parameters. The unit specification field uses a unit conversion matrix to unify the numerical values to a standard unit system. The time information field extracts minute, hour and day level time series features through a multi-scale time convolution network, and combines position embedding to mark the time series deviation of the collection time and the synchronization timestamp. After independent coding, the embedding vectors of each field are compressed in dimension through a fully connected layer, and are weighted and fused using an attention mechanism to generate a unified vector containing multi-dimensional information.

[0106] Specifically, the coding process of the data source identification field inputs the device number and the geographic location code into the pre-trained word embedding layer to generate a high-dimensional vector representing the spatial distribution of the device in the process chain. The data type field converts parameter labels into clustered representations in vector space through a semantic embedding model to distinguish the semantic boundaries of different physical properties. The unit specification field uses a preset conversion coefficient matrix to map the original numerical value and unit combination to a standardized numerical value without units. The time information field sets different convolution kernel sizes in parallel in the time convolution layer to extract short-term fluctuations and long-term trend features, and superimposes them with the position coding of the timestamp. After normalization, each subspace vector calculates the weight coefficient through cross-channel attention, dynamically adjusts the fusion ratio, and finally outputs a composite vector that preserves the physical semantic association, process position topology and time evolution law, providing a multi-dimensional feature basis for knowledge graph construction.

[0107] As a preferred embodiment, the scheme of the application is implemented as follows: in the implementation process of a certain sewage treatment plant, a full-process information knowledge graph is constructed based on the pretreated vectorized expression. In specific implementation, the biological reaction tank process unit entity is mapped to a process node, the dissolved oxygen sensor device entity is mapped to a device node, the dissolved oxygen concentration monitoring parameter entity is mapped to a parameter node, and the running time period of 10:00-11:00 in the morning is mapped to a time segment node. According to the process logic between the biological reaction tank and the secondary sedimentation tank, a directed edge relationship between the process unit nodes is established, and the directed edge attribute is set to "flow to the next processing stage". The "deployed in" relationship edge is established between the dissolved oxygen sensor node and the biological reaction tank process node, and the device installation time is 2023 May. When the dissolved oxygen parameter node abnormally fluctuates, the "occurs in" relationship edge is established between the parameter node and the time segment node, and the water load working condition of the day is marked.

[0108] Through the above technical scheme, the application effectively solves the problem that multi-source heterogeneous data in the sewage treatment process is difficult to construct a unified knowledge expression system. By converting the process entities and their associated relationships into a structured graph model, the spatial coupling expression of the monitoring parameter dynamic change and the process flow is realized. The time stamp attribute and the working condition attribute embedded in the knowledge graph enable the running anomaly to be accurately traced back to a specific processing stage and a time interval, provide an inferable causal relationship path for the intelligent early warning system, and improve the diagnosis efficiency of problems such as dissolved oxygen abnormality.

[0109] In some of the above schemes of the application, although the constructed information knowledge graph can represent the static association relationship of the full process of sewage treatment, it lacks the ability of dynamic state monitoring and abnormal propagation path tracing, cannot identify the parameter abnormal trend and locate the root node in real time, and leads to delayed early warning and difficulty in guiding precise regulation.

[0110] The application further proposes that after constructing the information knowledge graph for the full process of sewage treatment, the drift trend of the parameter node is identified according to the nodes and their associated edge relationships in the graph. The node state in the current graph is scored in real time according to the historical running data, and when the score result exceeds the threshold value, an early warning mark is triggered, the abnormal influence propagation path is determined through graph path backtracking, and early warning is performed according to the abnormal influence propagation path.

[0111] The dynamic state update includes updating the running parameter values of the process unit nodes, the real-time detection values of the sensor nodes, and the state marks of the time slice nodes at a minute level frequency, and synchronously updating the time stamp attributes and the working condition attribute on the associated edges. The parameter drift trend identification calculates the mean offset and the standard deviation change rate of the node parameters using a sliding window mechanism, and determines that the drift state occurs when the offset of three consecutive windows exceeds 15% of the historical baseline value. The real-time scoring is obtained by calculating the cosine similarity between the current parameter value and the historical data of the same period, and combining the state change rate of the upstream and downstream nodes of the process unit. The warning threshold is divided into three levels according to the process sensitivity, corresponding to yellow, orange and red alarm levels respectively. The graph path backtracking adopts a back propagation algorithm, traverses the influence path of the abnormal node in the reverse direction of the directed edge, and preferentially selects the path with the highest weight value as the main propagation link.

[0112] Specifically, the dynamic state update module continuously receives real-time detection data from the data acquisition layer, updates the current value of the secondary sedimentation tank dissolved oxygen concentration node to 4.2 mg / L, and synchronously modifies the time stamp attribute of the node and the dissolved oxygen control instruction edge of the biological reaction tank. The parameter drift identification module detects that the mean value of the disinfection tank residual chlorine concentration node rises from 0.3 mg / L to 0.52 mg / L in three consecutive 5-minute windows, and the standard deviation expands to 0.12, triggering the drift mark. The real-time scoring module calculates the similarity score between the current value of the sludge return ratio node and the historical data to be 0.63, and generates a comprehensive score of 82 points by combining the change rate of the upstream MLSS concentration node. When the score exceeds the preset threshold of 80 points, the warning module activates the orange alarm, and finds the abnormal propagation path through graph path backtracking, which is the sudden change of the influent flow → the fluctuation of the grid pool liquid level → the abnormal DO of the biological reaction tank → the decrease of the sludge settling efficiency of the secondary sedimentation tank. The operation and maintenance personnel check the influent flow meter and the grid control valve accordingly, so as to quickly locate and dispose the abnormal source.

[0113] As a preferred embodiment, the scheme of the application is implemented as follows: after constructing the information knowledge graph of the whole process of sewage treatment, the operation process of the real-time warning function includes: when the sludge dewatering machine running current value deviates from the normal fluctuation interval for three consecutive synchronization periods, the current monitoring node in the knowledge graph will trigger state update. The system automatically extracts the current value from the corresponding current-torque correlation mode in the historical operation data, calculates the similarity score between the current value and the historical mode library. If the similarity score is lower than the preset threshold, the warning module will activate the path backtracking algorithm, trace the dissolved oxygen abnormal node of the biological reaction tank in the reverse direction of the upstream and downstream relationship of the process unit in the knowledge graph, and identify the abnormal attribute of the associated edge between the sludge concentration sensor and the dosing pump control instruction. The operation and maintenance personnel can quickly locate the mechanical failure of the aeration amount regulating valve of the biological reaction tank according to the abnormal propagation path report generated by the system.

[0114] By the technical solution, the application realizes accurate tracing and active early warning of abnormal working conditions in the sewage treatment process, effectively solves the problem that the traditional monitoring system cannot identify multi-parameter associated abnormalities. Through the dynamic node updating mechanism of the knowledge graph, the early trend change of parameter drift can be captured. Based on the backtracking analysis of the graph structure path, the propagation path of the abnormality in the process chain can be automatically restored, avoiding the low efficiency problem of manual troubleshooting. The real-time scoring mechanism combined with the historical data mode library improves the accuracy of early warning judgment and prevents false positives and false negatives.

[0115] In the above embodiments, by arranging distributed sensors at each processing stage and adopting a unified access protocol and timing synchronization mechanism, spatiotemporal consistent collection of raw data is realized, label normalization and standardization are completed through semantic matching, then metadata features are extracted through structured coding, and a data cross-layer representation model based on transfer learning is constructed, realizing unified vectorization expression of time sequence, semantic and structural features between different data sources. Based on the vector representation, an information knowledge graph oriented to the whole process is constructed, improving the data fusion, interpretability and usability, breaking the traditional data island pattern, and providing unified data support for intelligent analysis, abnormal early warning, process optimization and other core functions of smart water.

[0116] Those skilled in the art should understand that the embodiments of the application can be provided as a method, system or computer program product. Therefore, the application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0117] The application is described with reference to flowcharts and / or block diagrams according to the method, device (system) and computer program product of the embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a machine that implements the function specified in the flowcharts and / or block diagrams. Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The device that realizes the function specified in one block or multiple blocks.

[0118] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The functions specified in the flow or flows and / or blocks Figure 1 The functions specified in the flow or flows and / or blocks

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow Figure 1 The functions specified in the flow or flows and / or blocks Figure 1 The functions specified in the flow or flows and / or blocks

[0120] Finally, it should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limiting the same. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and any modification or replacement without departing from the spirit and scope of the present application should be included in the protection scope of the claims of the present application.< / hasunit> < / sensor001>

Claims

1. A multi-source information fusion heterogeneous data regularization method, characterized in that, The application comprises the following steps: Distributed sensors are arranged at several treatment stages of the sewage treatment process, and detection data of all the distributed sensors are collected synchronously through an access protocol and time sequence; The detection data are semantically analyzed based on semantic matching, and the detection data are normalized and standardized in terms of semantic labels; The detection data after semantic standardization are processed to obtain metadata containing data source identification, data type, unit specification, and time information; A data cross-layer representation model based on transfer learning is constructed, and feature extraction and semantic embedding of the metadata are performed according to the data cross-layer representation model to obtain a unified vectorized expression; An information knowledge graph for the whole process of sewage treatment is constructed based on the vectorized expression.

2. The method of claim 1, wherein, When arranging distributed sensors at several treatment stages of the sewage treatment process, the following steps are included: Sensor nodes with temperature, dissolved oxygen, pH, conductivity, turbidity, flow, liquid level, chemical oxygen demand, and ammonia nitrogen detection functions are arranged at the grid pool, sand settling pool, biological reaction pool, secondary sedimentation tank, disinfection tank, and sludge treatment stage.

3. The method of claim 1, wherein, When collecting the detection data of all the distributed sensors synchronously through an access protocol and time sequence, the following steps are included: Various sensors are connected to the same data acquisition network through heterogeneous communication protocols such as OPC-UA, MQTT, or HTTP protocol, and unified data packaging and format standardization are performed through protocol adaptation middleware; The distributed sensors are synchronized in time through the network time protocol or IEEE 1588 precision time protocol.

4. The method of claim 1, wherein, When the detection data are semantically analyzed based on semantic matching, and the detection data are normalized and standardized in terms of semantic labels, the following steps are included: The original description information of the detection data is subjected to semantic feature extraction through natural language processing, which includes keyword recognition, part-of-speech analysis, entity extraction, and semantic nested analysis; Based on vector semantic similarity calculation, semantic distance measurement, and upper and lower semantic relationship matching, the detection data are mapped to corresponding standard label items for the normalization of synonymous, polysemous, and abbreviated data labels; The normalized labels are associated with corresponding physical meanings, unit systems, and belonging processing units to output structured semantic label items.

5. The method of claim 4, wherein, When the detection data after semantic standardization are processed to obtain metadata containing data source identification, data type, unit specification, and time information, the following steps are included: The detection data after semantic standardization are field-decomposed according to a preset structure template to extract and collect at least the following: Data source identification field for recording sensor device number, belonging processing unit, and geographic location code; Data type field for indicating the physical property type of the detection value; Unit specification field for unifying the physical unit corresponding to the numerical value; Time information field for recording the original collection time, system reception time, and synchronization time stamp; All fields are packaged as structured metadata items based on JSON, XML, or RDF semantic format representation, and the original data mapping index of the metadata is preserved.

6. The method of claim 1, wherein, When constructing a data cross-layer representation model based on transfer learning, the following steps are included: A dual-branch hybrid model architecture is adopted, which is composed of a pre-trained semantic encoder and a graph neural network, the pre-trained semantic encoder is used to process structured semantic labels contained in the metadata to generate initial semantic embedding vectors; The graph neural network takes multiple wastewater treatment process units and their upstream and downstream relationships as nodes and edges to establish a process level structure graph, and based on the input metadata, feature propagation and aggregation are performed on the graph to obtain structured embedding vectors.

7. The method of claim 6, wherein, When constructing a data cross-layer representation model based on transfer learning, it also includes: Through source domain training in wastewater treatment projects that have been semantically annotated and process structure mapped, initial model parameters are obtained; Based on the domain adaptation transfer strategy, the model is transferred to the target wastewater treatment scene, and adjusted through label samples; The domain adaptation transfer strategy process includes feature distribution alignment, maximum mean difference regularization and semantic label consistency constraint.

8. The method of claim 7, wherein, When the data cross-layer representation model is used to extract features and embed semantics from the metadata to obtain a unified vector representation, it includes: The data source identification field, data type field, unit specification field and time information field in the metadata are independently subspace encoded, and word embedding, position embedding and multi-scale time convolution network are used to extract features and model time series from each field; After fusing the embedding vectors of each field, dimension alignment and feature compression are performed to obtain the vector representation, which contains information substructures for representing physical semantics, processing stage position and dynamic change trend.

9. The method of claim 8, wherein, When the vector representation is used to construct an information knowledge graph for the whole process of wastewater treatment, it includes: The entity elements in the vector representation are mapped to nodes in the knowledge graph, including process units, sensor devices, monitoring parameters, time segments and control instructions; According to the wastewater treatment process logic and process structure, the directed edge relationship between nodes is established, and the timestamp or working condition is added as the attribute of the edge; The information knowledge graph for the whole process of wastewater treatment is obtained.

10. The method of claim 9, wherein, After the vector representation is used to construct an information knowledge graph for the whole process of wastewater treatment, it also includes: According to the information knowledge graph for the whole process of wastewater treatment, an early warning is performed; The dynamic state of each node and its associated edge relationship in the graph is updated to identify the drift trend of the parameter node; According to the historical operation data, the node state in the current graph is scored in real time, and when the score result exceeds the threshold, an early warning mark is triggered, the abnormal influence propagation path is determined through graph path backtracking, and the early warning is performed according to the abnormal influence propagation path.

Images