A multi-source data semantic interaction method and system for the water industry

Abstract

The application belongs to the technical field of water industry, and relates to a multi-source data semantic interaction method and system for the water industry, comprising: analyzing and processing multi-source monitoring data collected in real time to obtain a structured semantic data set containing multiple parameters; based on a preset early warning threshold, performing abnormality judgment on a target parameter; the target parameter is a specific monitoring parameter selected in the structured semantic data set; when the target parameter is abnormal, based on a pre-constructed correlation retrieval fusion model and the structured semantic data set, calculating the causal correlation strength of a first-level correlation parameter directly correlated with the abnormal target parameter; and according to the causal correlation strength of the first-level correlation parameter, the structured semantic data set and a parameter correlation rule, dynamically reasoning the abnormal reason of the abnormal target parameter. The application breaks through the semantic barrier of multi-source monitoring data, improves the accuracy of alarm, can accurately locate the abnormal reason, realizes efficient decision-making, and adapts to the multi-scene requirements of the water industry.

Description

Technical Field

[0001] This invention relates to the technical field of the water industry, and in particular to a multi-source data semantic interaction method and system for the water industry. Background Technology

[0002] In the daily operations of water utilities, a large amount of data from different process systems is involved, such as water quality data, process operation data, and equipment monitoring status data. Achieving intelligent analysis of multi-source data and providing early warnings of anomalies is key to ensuring wastewater treatment, water supply security, improving operational efficiency, and reducing energy consumption.

[0003] However, existing technologies still have significant shortcomings in anomaly monitoring of water systems. On the one hand, the lack of a unified standard for data semantics makes it difficult to correlate multi-source data. For example, COD data may come from online monitoring instruments, expressed as CODcr in mg / L; COD data may also come from manual records, such as recording chemical oxygen demand as 45; the PLC control system does not provide an interpretation for its COD_01 field, making it impossible to determine the type or unit of the data in this field. On the other hand, anomaly monitoring suffers from slow location, high false alarm rates, and an inability to accurately analyze the causes of anomalies. For example, when a COD concentration is detected to be greater than 50 mg / L, although an alarm for exceeding the COD limit will be triggered quickly, the cause of the alarm cannot be analyzed. When the COD concentration exceeds the limit, manual troubleshooting by operators is still required, which is time-consuming and prone to missing key information. Summary of the Invention

[0004] Based on the above analysis, the embodiments of the present invention aim to provide a multi-source data semantic interaction method and system for the water industry, in order to solve the problems of inconsistent semantics, low efficiency of multi-source data association, lack of semantic support for abnormal alarms, and inability to accurately locate the cause of abnormalities in the existing technology.

[0005] On one hand, embodiments of the present invention provide a multi-source data semantic interaction method for the water industry, including:

[0006] Based on a pre-built semantic ontology library, real-time collected multi-source monitoring data is parsed and processed to obtain a structured semantic dataset containing multiple parameters; the relationship between each parameter is determined based on the parameter association rules in the semantic ontology library.

[0007] Based on a preset warning threshold, anomaly detection is performed on the target parameter; the target parameter is a specific monitoring parameter selected from the structured semantic dataset.

[0008] When the target parameter is abnormal, the causal association strength of the first-level association parameters directly associated with the abnormal target parameter is calculated based on the pre-built association retrieval fusion model and the structured semantic dataset.

[0009] Based on the causal correlation strength of the first-level correlation parameters, the structured semantic dataset, and the parameter correlation rules, the abnormal cause of the abnormal target parameter is dynamically inferred.

[0010] Furthermore, based on a pre-built semantic ontology library, the real-time collected multi-source monitoring data is parsed and processed to obtain a structured semantic dataset containing multiple parameters, including:

[0011] Based on the semantic parsing rules in the semantic ontology library, the multi-source monitoring data is preprocessed by performing semantic consistency verification, semantic completion, and abnormal semantic marking.

[0012] Based on the preprocessing results, multiple parameters and the structured semantic information corresponding to each parameter are determined, and the structured semantic dataset is constructed.

[0013] Furthermore, the calculation of the causal association strength of the first-level association parameters directly associated with the abnormal target parameters based on the pre-built association retrieval fusion model includes:

[0014] Determine the abnormal time period of the abnormal target parameter;

[0015] Extract the full time-series data of the abnormal target parameters and the first-level correlation parameters during the abnormal period from the structured semantic dataset;

[0016] The full time-series data is used as input to the association retrieval fusion model, and the causal association strength of the first-level association parameter is calculated by combining the parameter association rules.

[0017] Furthermore, the association retrieval fusion model includes at least: an adaptive feature extraction layer for performing feature extraction, a nonlinear association modeling layer for performing association strength calculation, and a dynamic causal verification layer for performing causal verification.

[0018] Furthermore, the calculation process for the causal correlation strength of the first-level correlation parameter specifically includes:

[0019] Based on the adaptive feature extraction layer, adaptive feature extraction and sub-band filtering are performed on the full time series data to obtain the feature sub-bands corresponding to each parameter in the full time series data, and the effective duration of the abnormal target parameter being affected by each first-level correlation parameter is determined.

[0020] Based on the nonlinear correlation modeling layer, the nonlinear correlation strength between the abnormal target parameter and each first-level correlation parameter is calculated according to the feature sub-bands corresponding to each parameter.

[0021] Based on the dynamic causal verification layer, the nonlinear correlation strength is dynamically verified by utilizing the effective duration of the influence of each first-level correlation parameter on the abnormal target parameter; based on the result of the dynamic causal verification, the causal correlation strength between the abnormal target parameter and each first-level correlation parameter is obtained.

[0022] Further, the step of dynamically inferring the abnormal cause of the abnormal target parameter based on the causal association strength of the first-level association parameter, the structured semantic dataset, and the parameter association rules includes:

[0023] Compare the causal correlation strength of each of the first-level correlation parameters, and determine the real-time monitoring value corresponding to the first-level correlation parameter with the highest causal correlation strength;

[0024] From the structured semantic dataset, retrieve the real-time monitoring values ​​of the secondary association parameters that are directly associated with the primary association parameters during the abnormal period;

[0025] Based on the parameter association rules, and according to the real-time monitoring values ​​of the first-level association parameter with the highest causal association strength and the second-level association parameters, the abnormal cause of the abnormal target parameter is dynamically inferred.

[0026] Furthermore, the method also includes:

[0027] Based on the cause of the abnormality of the abnormal target parameter, an alarm message is generated, and combined with the historical case library, a handling suggestion is generated for the cause of the abnormality.

[0028] On the other hand, embodiments of the present invention provide a multi-source data semantic interaction system for the water industry, including:

[0029] The semantic engine unit is used to build a semantic ontology library;

[0030] The semantic parsing unit is used to parse and process the real-time collected multi-source monitoring data based on the semantic parsing rules to obtain a structured semantic dataset containing multiple parameters; wherein the relationship between each parameter is determined based on the parameter association rules in the semantic ontology library.

[0031] An anomaly monitoring unit is used to determine anomalies in target parameters based on a preset warning threshold; the target parameters are specific monitoring parameters selected from the structured semantic dataset.

[0032] The association retrieval unit is used to calculate the causal association strength of the first-level association parameters directly associated with the abnormal target parameter based on the pre-built association retrieval fusion model and the structured semantic dataset when the target parameter is abnormal.

[0033] The dynamic reasoning unit is used to dynamically infer the abnormal cause of the abnormal target parameter based on the causal association strength of the first-level association parameter, the structured semantic dataset, and the parameter association rules.

[0034] Furthermore, the associated retrieval unit is also used for:

[0035] Determine the abnormal time period of the abnormal target parameter;

[0036] Extract the full time-series data of the abnormal target parameters and the first-level correlation parameters during the abnormal period from the structured semantic dataset;

[0037] The full time-series data is used as input to the association retrieval fusion model, and the causal association strength of the first-level association parameter is calculated by combining the parameter association rules.

[0038] Furthermore, the dynamic reasoning unit is also used for:

[0039] Compare the causal correlation strength of each of the first-level correlation parameters, and determine the real-time monitoring value corresponding to the first-level correlation parameter with the highest causal correlation strength;

[0040] From the structured semantic dataset, retrieve the real-time monitoring values ​​of the secondary association parameters that are directly associated with the primary association parameters during the abnormal period;

[0041] Based on the parameter association rules, and according to the real-time monitoring values ​​of the first-level association parameter with the highest causal association strength and the second-level association parameters, the abnormal cause of the abnormal target parameter is dynamically inferred.

[0042] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0043] First, unlike the problem of inconsistent semantics in multi-source data in related technologies, this invention transforms multi-source core data in the water industry into a structured semantic dataset of "entity-attribute-relationship" by constructing a semantic ontology library. It clearly defines the process-specific semantic relationships between multi-source parameters. At the same time, by assigning a unique identifier to each type of data, it can associate corresponding full-dimensional semantic information and automatically parse unfamiliar identifiers into standard semantics, thus breaking through the semantic barriers of multi-source data.

[0044] Second, unlike the low efficiency of multi-source data association in related technologies, this invention can perform semantic association retrieval in real time. Through multi-scale feature extraction, nonlinear modeling and dynamic causal verification, it achieves deep integration of "data features - process semantics - dynamic association". It can quickly associate abnormal parameters with contemporaneous data, providing strong support for locating the cause of abnormalities.

[0045] Third, unlike related technologies that cannot accurately locate the cause of anomalies, this invention constructs a logical chain from data features to cause reasoning through a process semantic reasoning algorithm. It can not only dynamically reason the cause of parameter anomalies, but also form an integrated closed-loop system of "alarm-cause location-decision suggestion", realizing accurate alarm and efficient decision-making. At the same time, it can be quickly extended to semantic interaction of various water quality indicators, adapting to the multi-scenario needs of the water industry.

[0046] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0047] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0048] Figure 1 This is a flowchart of a multi-source data semantic interaction method for the water industry according to an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram illustrating the reasoning logic of a multi-source data semantic interaction method for the water industry according to an embodiment of the present invention.

[0050] Figure 3 This is a semantic association graph between the parameters in the embodiments of the present invention;

[0051] Figure 4 This is a schematic diagram of the main modules of a multi-source data semantic interaction system for the water industry according to an embodiment of the present invention. Detailed Implementation

[0052] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0053] A specific embodiment of the present invention discloses a multi-source data semantic interaction method for the water industry, such as... Figure 1 As shown, the steps S1 to S4 are as follows:

[0054] Step S1: Based on the pre-built semantic ontology library, the real-time collected multi-source monitoring data is parsed and processed to obtain a structured semantic dataset containing multiple parameters; wherein the relationship between each parameter is determined based on the parameter association rules in the semantic ontology library.

[0055] In implementation, the semantic ontology library can be constructed based on a template structure of parameter entities, entity attributes, and entity relationships; the semantic ontology library includes at least semantic parsing rules and parameter association rules. First, various core parameters within the industry domain are designated as parameter entities; for example, COD concentration (chemical oxygen demand), DO value (dissolved oxygen), aeration intensity, and dosage. Second, each parameter entity is configured with entity attributes describing its characteristics, such as data units, dimensions, warning thresholds, and normal value ranges. Then, the mutual influence or dependency relationships between different parameter entities are designated as entity relationships. In other words, the association relationships between various parameters, such as target parameters, first-level associated parameters, and second-level associated parameters, in subsequent steps of this invention are all determined based on the parameter association rules in the semantic ontology library.

[0056] Furthermore, based on the defined parameter entities and entity attributes, semantic parsing rules are generated. These rules are used to identify and semantically annotate the input multi-source monitoring data, parsing it into structured data with unified semantics. Then, based on the defined entity relationships, parameter association rules are determined. These association rules are the process-specific semantic relationships between parameters. For example, "COD concentration ↑ → DO value ↓" indicates that an increase in COD concentration is associated with a decrease in DO value; "dosage ↑ → COD concentration ↓" indicates that an increase in dosing is associated with a decrease in COD concentration; and "DO value ↓ → aeration intensity ↓" indicates that a decrease in DO value is associated with a decrease in aeration intensity.

[0057] As a result, a semantic system specific to the water industry was constructed, enabling semantic interaction and rapid association between multi-source data from different systems, breaking through the semantic barriers of multi-source data, and providing a foundation for subsequent real-time data analysis.

[0058] During the data acquisition and preprocessing stage, multiple data acquisition systems can be used to collect multi-source monitoring data in the wastewater treatment plant in real time and store it in a hybrid storage system consisting of relational databases and NoSQL databases, forming a unified data list containing fields such as time, data_type, data_value, unit, device_id, and data_source.

[0059] For example, the core data and its fields in a hybrid storage system mainly include:

[0060] (1) Water quality monitoring data: COD concentration (cod_value, unit mg / L), DO value (do_value, unit mg / L), and the monitoring time (time) and equipment number (device_id) for each parameter;

[0061] (2) Process operation data: aeration intensity (m3 / h), dosage (L / h), etc.

[0062] (3) Equipment status data: monitor calibration status (cal_status), fault log (fault_log, unstructured text), etc.;

[0063] (4) Records of parameters exceeding the standard in the past 12 months and their corresponding process adjustment plans, such as COD exceeding the standard record and process adjustment plan.

[0064] The data semantic parsing stage specifically includes: performing preprocessing operations such as semantic consistency verification, semantic completion, and abnormal semantic marking on the multi-source monitoring data based on the semantic parsing rules in the semantic ontology library.

[0065] The semantic consistency verification includes: checking whether the semantics of the data match the semantic ontology library, for example, automatically converting the unit "g / L" of COD concentration to "mg / L" and marking "Unit conversion: 1g / L = 1000mg / L"; semantic completion includes: completing missing semantic data, for example, associating the semantic ontology library with the equipment number and completing "unit: mg / L" for "DO = 1.8"; and abnormal semantic marking includes: marking obviously abnormal data, for example, marking "suspected equipment failure, needs to be verified in conjunction with equipment status logs", etc., to avoid false alarms.

[0066] Subsequently, based on the preprocessing results, multiple parameters and the structured semantic information corresponding to each parameter are determined, and the structured semantic dataset is constructed. For example, the fault log "COD instrument reading fluctuation at 14:00" is parsed into the structured semantic information "Data type: equipment abnormality; associated data: COD monitor; time: 14:00; abnormality type: reading fluctuation".

[0067] Preferably, during the analysis process, a unique semantic identifier needs to be assigned to each type of parameter in the multi-source monitoring data, and a unified semantic identifier system needs to be constructed. For example, the identifier "SW-COD-001" is assigned to the COD concentration parameter. This unique semantic identifier can be used as a parameter index to associate standardized semantic information of this type of parameter across all dimensions.

[0068] For example, semantic identifiers can be used to associate parameters with their Chinese names, standard units of measurement, threshold ranges, etc. Based on this system, when parsing raw data fields from different systems, the system can automatically map unfamiliar identifiers (such as the "COD_01" field in a PLC control system) to their corresponding unique semantic identifiers (such as SW-COD-001) through matching rules. This parses the data into standard data that is globally recognizable by the system, achieving the effect of one-time registration and semantic interoperability across the entire system. This ensures the semantic consistency of data at each stage of acquisition, storage, and analysis, providing a reliable foundation for subsequent association retrieval and analysis.

[0069] Step S2: Based on a preset warning threshold, anomaly judgment is made on the target parameter; the target parameter is a specific monitoring parameter selected in the structured semantic dataset.

[0070] The key parameters such as COD concentration, DO value and aeration intensity in the embodiments of the present invention are structured monitoring parameters obtained by extracting and standardizing multi-source monitoring data such as water quality monitoring data, process operation data and equipment status data through semantic parsing rules.

[0071] It's understandable that dissolved oxygen (DO) is one of the core parameters in biological treatment processes at wastewater treatment plants. Chemical oxygen demand (COD) refers to the amount of oxygen consumed by strong oxidants in the oxidation of organic matter in a water sample under specific conditions, usually expressed in mg / L. The potassium dichromate method is generally used for COD determination; therefore, COD is also recorded as CODcr. A higher COD value indicates more severe organic pollution in the water and poorer water quality. In wastewater treatment, COD is one of the important indicators for assessing water quality, reflecting the degree of pollution and self-purification capacity of the water body.

[0072] Of course, the present invention is highly scalable and can be adapted to the needs of multiple scenarios in the water industry. It can also monitor other water quality indicators such as ammonia nitrogen and total phosphorus.

[0073] In actual monitoring, one or more specific parameters can be selected as target parameters for continuous monitoring and anomaly detection. Preferably, a semantic ontology library can be invoked to extract standard alarm thresholds bound to semantic identifiers of the target parameters. When the real-time monitored value exceeds a preset multiple of this threshold, the water quality data is determined to be out of standard and an anomaly is triggered; conversely, if the value is below the preset multiple, it is determined to be normal. For example, COD concentration can be used as a target parameter, and combined with a preset warning threshold in the semantic ontology library, the semantic definition "CODcr, exceeding warning threshold 45mg / L" can be automatically associated with it. The real-time monitored value of COD concentration in the structured semantic dataset can be compared with a multiple of this warning threshold to determine whether the COD concentration is in an abnormal state of exceeding the standard.

[0074] In addition, when a user sets a new warning threshold, the system can obtain the new threshold and automatically update it to the semantic information field of the corresponding semantic identifier in the semantic ontology library, such as updating the "alarm threshold" field, thereby realizing centralized maintenance and immediate effect of the threshold.

[0075] Step S3: When the target parameter is abnormal, the causal association strength of the first-level association parameters directly associated with the abnormal target parameter is calculated based on the pre-built association retrieval fusion model and the structured semantic dataset.

[0076] It is understandable that when monitoring for anomalies in a target parameter, there may be multiple first-level related parameters of the same type in the structured semantic dataset. First-level related parameters refer to parameters directly associated with the target parameter, as determined by parameter association rules. Therefore, it is necessary to perform deep association retrieval on the structured semantic dataset to determine the unique related parameter most relevant to the current anomaly scenario from among multiple directly related parameters of the same type, in order to accurately obtain its true real-time monitoring value and thus accurately pinpoint the cause of the anomaly.

[0077] In practice, the specific steps include: determining the abnormal time period of the abnormal target parameter; extracting the full time-series data of the abnormal target parameter and the first-level association parameter within the abnormal time period from the structured semantic dataset; using the full time-series data as input to the association retrieval fusion model, and calculating the causal association strength of the first-level association parameter in conjunction with the parameter association rules.

[0078] Preferably, the association retrieval fusion model includes at least: an adaptive feature extraction layer for performing feature extraction, a nonlinear association modeling layer for performing association strength calculation, and a dynamic causal verification layer for performing causal verification.

[0079] For example, the adaptive feature extraction layer can be implemented based on the adaptive wavelet packet transform (AWPT) technique, the linear correlation modeling layer can be implemented based on the kernel partial least squares (KPLS) technique, and the dynamic causality verification layer can be implemented based on the dynamic Granger causality test (DGCT) technique.

[0080] Preferably, the calculation process of the causal correlation strength of the first-level correlation parameter specifically includes:

[0081] Based on the adaptive feature extraction layer, adaptive feature extraction and sub-band filtering are performed on the full time series data to obtain the feature sub-bands corresponding to each parameter in the full time series data, and the effective duration of the abnormal target parameter being affected by each first-level correlation parameter is determined.

[0082] Based on the nonlinear correlation modeling layer, the nonlinear correlation strength between the abnormal target parameter and each first-level correlation parameter is calculated according to the feature sub-bands corresponding to each parameter.

[0083] Based on the dynamic causal verification layer, the nonlinear correlation strength is dynamically verified by utilizing the effective duration of the influence of each first-level correlation parameter on the abnormal target parameter; based on the result of the dynamic causal verification, the causal correlation strength between the abnormal target parameter and each first-level correlation parameter is obtained.

[0084] In some implementations, reference Figure 2 As shown, using COD concentration as the target parameter, a deep association retrieval of COD concentration and DO values ​​(i.e., first-level association parameters) in a structured semantic dataset can be achieved based on an association retrieval fusion model. That is, from multiple DO values ​​in the structured semantic dataset (such as DO1, DO2, DO3, etc.), the DO value with the strongest causal association is selected to determine the true state of the DO value.

[0085] Specifically, a certain period of time before and after the COD exceeding the standard (e.g., 14:00) is designated as an abnormal period (e.g., 12:00-14:30). According to a preset sampling step size (e.g., 1 minute), the full time series data within this abnormal period is extracted from the structured semantic dataset and used as the input of the model. This full time series data is the DO / COD data of multiple process segments.

[0086] First, the feature extraction process of the adaptive feature extraction layer specifically includes:

[0087] Semantic guidance: The parameter association rule of "DO value has a higher impact weight in the main process section" is called in the semantic ontology library. Combined with the general process experience of the water industry, the initial search range is set to 30-90 minutes to cover the common range of DO value affecting COD concentration.

[0088] Adaptive filtering: Using the information entropy criterion, the information gain of sub-bands in the 30 / 45 / 60 / 75 / 90 minute intervals is calculated, and the sub-bands with the minimum information entropy and the highest feature discrimination are retained to determine the effective duration of the influence of DO value on COD concentration under the current operating conditions.

[0089] Secondly, the calculation process of the association strength in the nonlinear association modeling layer specifically includes:

[0090] Combining the "DO value weighting rules for main / secondary process sections" in the semantic ontology library, for example, the DO1 value weight of the main process section is 0.7 and the DO2 value weight of the secondary process section is 0.3, the multi-DO value sub-band features after adaptive filtering are used as independent variables, and the COD concentration sub-band features are used as dependent variables.

[0091] The features are mapped to a high-dimensional space using the radial basis function (RBF). Principal components are extracted using partial least squares to eliminate multicollinearity of multiple DO values. The nonlinear correlation strength between each DO value and COD concentration is quantified. The correlation strength ranges from 0 to 1. When the correlation strength is not less than 0.6, it is determined to be a highly correlated DO value.

[0092] Finally, the causal verification process of the dynamic causal verification layer specifically includes:

[0093] Based on the effective duration determined by the adaptive screening process, the DGCT lag order range is automatically matched. For example, if the effective duration of the calculated DO value for COD concentration is 70 minutes, then the lag order is set to 70, with 1 minute as a unit order. That is, it examines whether the DO value in the past 70 minutes has a causal effect on the current COD concentration.

[0094] The significance test is performed using the F statistic in DGCT, that is, to test whether the nonlinear association strength between each DO value and COD concentration is real and significant. The probability P corresponding to the F statistic is compared with the significance criterion. For example, when P < 0.05, it is determined to be a strong causal relationship, excluding spurious associations with synchronous fluctuations but no causality. The output is the lag response time of DO value to COD concentration and the causal association strength.

[0095] Thus, a comprehensive output result is generated, including the ranking of causal correlation strength of multiple DO values ​​(including causal labeling), dynamic effective duration, and lag response time. Preferably, historical correlation cases of similar operating conditions can be retrieved simultaneously, such as "when the influent COD = 350 mg / L on 2024-09-01, the DO1 correlation degree is 0.91", etc., so as to realize the closed loop of process adjustment and effect verification in subsequent steps.

[0096] Step S4: Based on the causal association strength of the first-level association parameters, the structured semantic dataset, and the parameter association rules, dynamically infer the abnormal cause of the abnormal target parameter.

[0097] In practice, the specific steps include: comparing the causal correlation strength of each of the primary correlation parameters and determining the real-time monitoring value corresponding to the primary correlation parameter with the highest causal correlation strength; retrieving the real-time monitoring values ​​of secondary correlation parameters directly associated with the primary correlation parameters during the abnormal period from the structured semantic dataset; and dynamically inferring the abnormal cause of the abnormal target parameter based on the parameter correlation rules, according to the real-time monitoring values ​​of the primary correlation parameters with the highest causal correlation strength and the real-time monitoring values ​​of the secondary correlation parameters.

[0098] Continue to refer to Figure 2As shown, when the target parameter COD = 48 mg / L, which exceeds the COD warning threshold of 45 mg / L, and the DO value with the highest causal correlation strength is determined to be 1.8 mg / L, which is less than the DO warning threshold of 2 mg / L, an association retrieval is performed on the structured semantic dataset to find the real-time monitoring value of the aeration intensity (secondary association parameter) directly related to the DO value during the abnormal period. For example, the aeration intensity at 13:25 was found to have increased from 20 mg / L. 3 / h dropped to 15m 3 / h; After that, the parameter association rules in the semantic ontology library can be called to construct a logical chain from data features to cause reasoning. From the parameter association rules "COD concentration↑→DO value↓" and "DO value↓→aeration intensity↓", the logical chain "insufficient aeration intensity leads to low DO value, causing COD to exceed the standard" can be dynamically reasoned to determine that the abnormal cause of COD exceeding the standard is due to insufficient aeration intensity.

[0099] refer to Figure 3 As shown, in this embodiment, the primary correlation parameter (such as DO value) refers to the parameter that is directly related to the target parameter (such as COD concentration) based on the parameter correlation rule, and the secondary correlation parameter (such as aeration intensity) refers to the parameter that is directly related to the primary correlation parameter (such as DO value) based on the parameter correlation rule.

[0100] It can be understood that the first-level association parameters serve as intermediate parameters, while the second-level association parameters serve as causal parameters. Furthermore, the purpose of retrieving the second-level association parameters from the structured semantic dataset is to verify whether the actual changes in the anomaly target parameters, first-level association parameters, and second-level association parameters follow the causal logic chain constructed by the semantic ontology library.

[0101] When there is only one secondary association parameter, its real-time monitoring value during the abnormal period is directly obtained from the structured semantic dataset. When there are multiple secondary association parameters, the unique parameter with the highest causal association strength and its real-time monitoring value are determined through the association retrieval fusion model. The specific association retrieval process is similar to that for determining the primary association parameter, and will not be elaborated here.

[0102] Of course, the diagnostic logic based on the first-level and second-level correlation parameters can be further extended along the causal logic chain to the third-level, fourth-level and other multi-level correlation parameters to achieve the tracing of the cause of the anomaly.

[0103] Preferably, the method further includes: generating alarm information based on the cause of the abnormality of the abnormal target parameter, and generating handling suggestions for the cause of the abnormality in conjunction with a historical case library.

[0104] The handling recommendations include: trend characteristics of the target parameters, trend characteristics of related parameters, and operating instructions for relevant process equipment. For example, based on the aforementioned abnormal cause of "insufficient aeration intensity," corresponding alarm information and handling recommendations can be generated, specifically including: Alarm time: 2024-10-01 14:00; COD concentration: 48 mg / L; Exceedance range: 3 mg / L; Possible cause: Reduced aeration intensity (from 13:30 onwards, from 20 mg / L). 3 / h reduced to 15m 3 / h); Recommended measures: Adjust the aeration intensity to 22m 3 / h, continuously monitor changes in COD / DO parameters.

[0105] Meanwhile, based on the real-time monitoring values ​​of the parameters, DO value curves, aeration intensity curves, etc. can be generated. The alarm information and related curves can be sent to the semantic interactive display interface of the system and simultaneously sent to the mobile terminal of the operators, thereby improving the accuracy of alarms and the efficiency of decision-making, and shortening the processing time of abnormal response.

[0106] Preferably, the semantic interactive display interface of this embodiment of the invention can be divided into the following four functional areas:

[0107] Real-time data area: Used to display the real-time monitoring values ​​of various parameters in the structured semantic dataset; for example, displaying the real-time monitoring values ​​of COD concentration, DO value, and dosage, and marking the parameter exceeding the standard COD=48mg / L in red. At the same time, it displays its complete semantics "CODcr, unit mg / L, warning threshold 45mg / L".

[0108] Semantic Relationship Graph Area: This area uses a visual graph to display the semantic relationships between various parameters. For example, it shows the semantic relationship of "COD↑→DO↓→Aeration Intensity↓", and uses node size to represent the weight of the parameter's influence.

[0109] Alarm Information Area: Used to display alarm content and associate it with device status and historical cases.

[0110] Historical Trends Area: Used to display the change curves of various parameters; for example, the change curves of COD concentration and its related parameters over the past 24 hours, with nodes such as "aeration intensity decreased at 13:30, COD exceeded the standard at 14:00" marked with dashed lines.

[0111] As can be seen from the above, this invention is applicable to the semantic integration, real-time early warning and correlation analysis of multi-source monitoring data of sewage treatment plants, and realizes semantic interoperability of water data; especially for scenarios such as alarms for exceeding the standards of key water quality parameters, it can quickly locate the cause of the anomaly and provide early warning decision support.

[0112] It should be noted that this invention relates to data interaction technology for all business scenarios in the water industry (including water supply, sewage treatment, rainwater scheduling, pipeline operation and maintenance, and reclaimed water reuse). It is applicable to the semantic unification, correlation retrieval, and interactive decision-making of multi-source heterogeneous data (water quality, liquid level, pipeline pressure, flow rate, equipment operation, chemical dosing, user-end water pressure / water consumption, etc.) in the water industry. It belongs to the field of water informatization and data intelligence application and can support the refined operation and cross-business collaborative management of the entire water business chain (from raw water collection to terminal water supply, from sewage collection to compliant discharge).

[0113] To further illustrate the abnormal alarm process of the present invention, a specific embodiment is provided below.

[0114] Taking a COD concentration exceeding the standard alarm scenario at a municipal wastewater treatment plant (daily treatment capacity of 50,000 tons) as an example, the specific implementation process is as follows:

[0115] The first stage is data preparation.

[0116] The main sources of multi-source monitoring data include:

[0117] COD concentration online monitoring instrument: data format is JSON; fields are time, cod_value, device_id; semantic definition is CODcr, unit is mg / L;

[0118] Aeration system: Data format is CSV; fields are time, do_value, and aeration_intensity; semantic definition is DO value, unit mg / L, aeration intensity unit m. 3 / h;

[0119] Dosing system: sourced from a MySQL database; fields are time, drug_dosage, and drug_type; semantic definition is dosage, unit L / h, and drug type is polyacrylamide;

[0120] Device logs: sourced from a MongoDB database, with fields including time, device_id, status, and log_info;

[0121] Equipment status: Originates from the PLC control system, field is cal_status.

[0122] The second stage involves semantic interaction and dynamic reasoning.

[0123] (1) Identification registration: Assign the semantic identifier “SW-COD” to COD concentration, and associate it with “COD warning threshold 45mg / L, emission standard 50mg / L”; assign the semantic identifier “SW-DO” to DO value, and associate it with “DO normal range 2-4mg / L, negatively correlated with COD concentration”.

[0124] (2) Data preprocessing: The “COD data collected on 14:00 on 10:00 on 2025-10-01” was processed into “time:14:00, cod_value:48, device_id:COD-01, unit:mg / L”; semantic verification confirmed that the unit was correct and there were no missing fields; semantic consistency verification was performed on “DO1=1.7mg / L, DO2=2.9mg / L, DO3=3.1mg / L” to confirm that the unit was uniformly mg / L, and semantic completion was performed to confirm that the parameter validity met the standard.

[0125] (3) Related Search:

[0126] First, perform AWPT adaptive feature extraction, which includes:

[0127] Input data: Retrieve full time-series data from 12:00 to 14:30 before and after the COD concentration exceeded the standard at 14:00, including DO1, DO2 and DO3 values. Collect one data point per minute, for a total of 151 DO value data points, and simultaneously collect 151 COD concentration data points.

[0128] Semantic guidance: Based on the rule in the semantic ontology library that "DO value of the main process section has a higher impact weight", combined with the common process range in the water industry, the initial search range is set to 30-90 minutes.

[0129] Information entropy calculation: Information entropy was calculated for the DO / COD subband features in the intervals of 30 minutes (13:30-14:00), 45 minutes (13:15-14:00), 60 minutes (13:00-14:00), 70 minutes (12:50-14:00), and 90 minutes (12:30-14:00), respectively. The results were 0.42, 0.39, 0.38, 0.32, and 0.41.

[0130] Effective duration determination: The information entropy is the lowest (0.32) in the 70-minute interval and the feature discrimination is the highest. The effective duration of DO value for COD concentration under the current working condition is determined to be 70 minutes. The low-frequency sub-band features of DO1 / DO2 / COD in this interval are retained, and high-frequency random interference of 1 to 5 minutes is removed.

[0131] Secondly, the KPLS nonlinear correlation strength calculation is performed, specifically including:

[0132] Parameter configuration: Radial basis function (RBF) is used, with kernel parameter σ = 0.8. This kernel parameter is calibrated based on the factory's historical COD exceedance data for the past 3 months, and is assigned values ​​of DO1 = 0.7, DO2 = 0.3, and DO3 = 0.0 according to semantic weight rules.

[0133] Association strength calculation: The nonlinear association strength between each DO value and COD concentration was calculated using the KPLS model, where: DO1 = 0.83, DO2 = 0.64, DO3 = 0.37. DO1 and DO2 with an association strength of not less than 0.6 were selected to enter the causal verification stage.

[0134] Semantic correction: Based on the semantic ontology library rule that "DO < 2 mg / L has a significant impact on COD degradation", the association strength of DO1 = 1.7 mg / L (less than the DO warning threshold of 2 mg / L) was corrected. The corrected association strength of DO1 is: 0.83 × 1.1 = 0.91.

[0135] Finally, the DGCT dynamic causality verification is performed, which specifically includes:

[0136] Lag order matching: Based on the 70-minute effective duration determined in the AWPT stage, the lag order of DGCT is set to 70 (1 minute / order) to cover the complete response cycle of DO value to COD concentration.

[0137] Significance test: The F-statistic and probability P-value were calculated. The results showed that the P-value of DO1 was 0.008 (<0.05), indicating strong causality with a lag response time of 15 minutes; the P-value of DO2 was 0.09 (>0.05), indicating weak causality.

[0138] Related search results output: Generate a structured related report, including "DO1: relatedness 0.91, effective duration 70 minutes, lag 15 minutes, strong causality; DO2: relatedness 0.64, lag 20 minutes, weak causality; DO3 no relatedness".

[0139] (4) Semantic reasoning: Combining the results of the associated retrieval with the structured semantic dataset, "13:25 aeration intensity from 20m" 3 / h dropped to 15m 3 Based on the real-time monitoring value of " / h", the cause of the anomaly was deduced: insufficient aeration in process section 1 led to a persistently low DO1 level (DO value less than 2mg / L from 13:40) and a 15-minute delay, which in turn caused the COD concentration to exceed the standard at 14:00.

[0140] Alarm Generation: Generate alarm information and push it to the semantic interactive display interface. Simultaneously, retrieve historical cases such as "September 1, 2025: Similar COD concentration exceeded the standard; aeration intensity adjusted to 22m³ / h". 3 "After / h, and after 30 minutes, the COD concentration dropped to 42mg / L", and based on historical cases, treatment recommendations were generated.

[0141] The third stage is interface display and operation.

[0142] Specifically, this includes: after operators discover a red alarm on the display interface, they can view the aeration intensity change curve to confirm insufficient aeration after 13:30; and based on historical cases, determine a comprehensive handling plan, subsequently issuing an order to "adjust the aeration intensity to 22m". 3 The command " / h" will display the real-time trend of COD concentration and DO value. After 30 minutes, the COD concentration will drop to 43 mg / L and the alarm will be automatically deactivated.

[0143] It is understood that the above implementation methods are only for ease of understanding and simplification of description, and should not be construed as limiting the present invention. The present invention does not specifically limit the types of monitoring parameters, the construction method of the semantic ontology library, the association retrieval method, etc.

[0144] Therefore, it can be seen that the embodiments of the present invention can achieve one of the following beneficial effects:

[0145] First, it boasts strong semantic interoperability. It addresses the semantic differences in multi-source monitoring data within the water industry, standardizing the definitions and relationships of parameters such as COD concentration and DO value, thus avoiding data silos.

[0146] Second, the alarm accuracy is high. By using semantic reasoning to correlate process parameters, false alarms exceeding limits are reduced, and false data caused by equipment malfunctions are eliminated, improving the efficiency of cause location by 80%.

[0147] Third, it offers high decision-making efficiency. Historical case studies and suggested measures are provided simultaneously upon receiving an alarm, reducing processing time for operations personnel, improving efficiency, and ensuring system stability and reliability.

[0148] Fourth, it has strong scalability. In addition to monitoring parameters such as COD concentration, it can be quickly extended to semantic interaction of other water quality indicators such as ammonia nitrogen and total phosphorus, adapting to the needs of multiple scenarios in the water industry.

[0149] In another embodiment of the present invention, a multi-source data semantic interaction system for the water industry is proposed, such as... Figure 4 As shown, it specifically includes the following modules:

[0150] The semantic engine unit is used to build a semantic ontology library;

[0151] The semantic parsing unit is used to parse and process the real-time collected multi-source monitoring data based on the semantic ontology library to obtain a structured semantic dataset containing multiple parameters; wherein the relationship between each parameter is determined based on the parameter association rules in the semantic ontology library.

[0152] An anomaly monitoring unit is used to determine anomalies in target parameters based on a preset warning threshold; the target parameters are specific monitoring parameters selected from the structured semantic dataset.

[0153] The association retrieval unit is used to calculate the causal association strength of the first-level association parameters directly associated with the abnormal target parameter based on the pre-built association retrieval fusion model and the structured semantic dataset when the target parameter is abnormal.

[0154] The dynamic reasoning unit is used to dynamically infer the abnormal cause of the abnormal target parameter based on the causal association strength of the first-level association parameter, the structured semantic dataset, and the parameter association rules.

[0155] Preferably, the system further includes a cross-domain mapping unit for resolving semantic differences between different acquisition systems. For example, mapping the "aeration volume" of the aeration system to "aeration intensity, m" in a semantic ontology library. 3 " / h" ensures data semantic consistency.

[0156] Preferably, the associated retrieval unit is further used for:

[0157] Determine the abnormal time period of the abnormal target parameter;

[0158] Extract the full time-series data of the abnormal target parameters and the first-level correlation parameters during the abnormal period from the structured semantic dataset;

[0159] The full time-series data is used as input to the association retrieval fusion model, and the causal association strength of the first-level association parameter is calculated by combining the parameter association rules.

[0160] Preferably, the dynamic reasoning unit is further used for:

[0161] Compare the causal correlation strength of each of the first-level correlation parameters, and determine the real-time monitoring value corresponding to the first-level correlation parameter with the highest causal correlation strength;

[0162] From the structured semantic dataset, retrieve the real-time monitoring values ​​of the secondary association parameters that are directly associated with the primary association parameters during the abnormal period;

[0163] Based on the parameter association rules, and according to the real-time monitoring values ​​of the first-level association parameter with the highest causal association strength and the second-level association parameters, the abnormal cause of the abnormal target parameter is dynamically inferred.

[0164] The above-described method and system embodiments are based on the same principles, and their related aspects can be referenced from each other to achieve the same technical effects. For specific implementation processes, please refer to the foregoing embodiments, which will not be repeated here.

[0165] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0166] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-source data semantic interaction method for the water industry, characterized in that, include: Based on a pre-built semantic ontology library, the real-time collected multi-source monitoring data is parsed and processed to obtain a structured semantic dataset containing multiple parameters. The relationships between the parameters are determined based on the parameter association rules in the semantic ontology library; Based on a preset warning threshold, anomaly detection is performed on the target parameter; the target parameter is a specific monitoring parameter selected from the structured semantic dataset. When the target parameter is abnormal, the causal association strength of the first-level association parameters directly associated with the abnormal target parameter is calculated based on the pre-built association retrieval fusion model and the structured semantic dataset. Based on the causal correlation strength of the first-level correlation parameters, the structured semantic dataset, and the parameter correlation rules, the abnormal cause of the abnormal target parameter is dynamically inferred.

2. The semantic interaction method according to claim 1, characterized in that, The pre-built semantic ontology library is used to parse and process real-time collected multi-source monitoring data to obtain a structured semantic dataset containing multiple parameters, including: Based on the semantic parsing rules in the semantic ontology library, the multi-source monitoring data is preprocessed by performing semantic consistency verification, semantic completion, and abnormal semantic marking. Based on the preprocessing results, multiple parameters and the structured semantic information corresponding to each parameter are determined, and the structured semantic dataset is constructed.

3. The semantic interaction method according to claim 2, characterized in that, The pre-built association retrieval fusion model calculates the causal association strength of first-level association parameters directly related to the abnormal target parameters, including: Determine the abnormal time period of the abnormal target parameter; Extract the full time-series data of the abnormal target parameters and the first-level correlation parameters during the abnormal period from the structured semantic dataset; The full time-series data is used as input to the association retrieval fusion model, and the causal association strength of the first-level association parameter is calculated by combining the parameter association rules.

4. The semantic interaction method according to claim 3, characterized in that, The association retrieval fusion model includes at least: an adaptive feature extraction layer for performing feature extraction, a nonlinear association modeling layer for performing association strength calculation, and a dynamic causal verification layer for performing causal verification.

5. The semantic interaction method according to claim 4, characterized in that, The calculation process for the causal correlation strength of the first-level correlation parameter specifically includes: Based on the adaptive feature extraction layer, adaptive feature extraction and sub-band filtering are performed on the full time series data to obtain the feature sub-bands corresponding to each parameter in the full time series data, and the effective duration of the abnormal target parameter being affected by each first-level correlation parameter is determined. Based on the nonlinear correlation modeling layer, the nonlinear correlation strength between the abnormal target parameter and each first-level correlation parameter is calculated according to the feature sub-bands corresponding to each parameter. Based on the dynamic causal verification layer, the nonlinear correlation strength is dynamically verified by utilizing the effective duration of the influence of each first-level correlation parameter on the abnormal target parameter; based on the result of the dynamic causal verification, the causal correlation strength between the abnormal target parameter and each first-level correlation parameter is obtained.

6. The semantic interaction method according to claim 5, characterized in that, The step of dynamically inferring the cause of the abnormality of the abnormal target parameter based on the causal association strength of the first-level association parameter, the structured semantic dataset, and the parameter association rules includes: Compare the causal correlation strength of each of the first-level correlation parameters, and determine the real-time monitoring value corresponding to the first-level correlation parameter with the highest causal correlation strength; From the structured semantic dataset, retrieve the real-time monitoring values ​​of the secondary association parameters that are directly associated with the primary association parameters during the abnormal period; Based on the parameter association rules, and according to the real-time monitoring values ​​of the first-level association parameter with the highest causal association strength and the second-level association parameters, the abnormal cause of the abnormal target parameter is dynamically inferred.

7. The semantic interaction method according to claim 6, characterized in that, The method further includes: Based on the cause of the abnormality of the abnormal target parameter, an alarm message is generated, and combined with the historical case library, a handling suggestion is generated for the cause of the abnormality.

8. A multi-source data semantic interaction system for the water industry, characterized in that, include: The semantic engine unit is used to build a semantic ontology library; The semantic parsing unit is used to parse and process the real-time collected multi-source monitoring data based on the semantic ontology library to obtain a structured semantic dataset containing multiple parameters; wherein the relationship between each parameter is determined based on the parameter association rules in the semantic ontology library. An anomaly monitoring unit is used to determine anomalies in target parameters based on preset warning thresholds; The target parameter is a specific monitoring parameter selected from the structured semantic dataset; The association retrieval unit is used to calculate the causal association strength of the first-level association parameters directly associated with the abnormal target parameter based on the pre-built association retrieval fusion model and the structured semantic dataset when the target parameter is abnormal. The dynamic reasoning unit is used to dynamically infer the abnormal cause of the abnormal target parameter based on the causal association strength of the first-level association parameter, the structured semantic dataset, and the parameter association rules.

9. The semantic interaction system according to claim 8, characterized in that, The associated retrieval unit is also used for: Determine the abnormal time period of the abnormal target parameter; Extract the full time-series data of the abnormal target parameters and the first-level correlation parameters during the abnormal period from the structured semantic dataset; The full time-series data is used as input to the association retrieval fusion model, and the causal association strength of the first-level association parameter is calculated by combining the parameter association rules.

10. The semantic interaction system according to claim 9, characterized in that, The dynamic reasoning unit is also used for: Compare the causal correlation strength of each of the first-level correlation parameters, and determine the real-time monitoring value corresponding to the first-level correlation parameter with the highest causal correlation strength; From the structured semantic dataset, retrieve the real-time monitoring values ​​of the secondary association parameters that are directly associated with the primary association parameters during the abnormal period; Based on the parameter association rules, and according to the real-time monitoring values ​​of the first-level association parameter with the highest causal association strength and the second-level association parameters, the abnormal cause of the abnormal target parameter is dynamically inferred.

Images