Method, device and equipment for identifying running state of equipment of pollution discharge enterprise

Abstract

The invention relates to a pollution discharge enterprise equipment operation state identification method, device and equipment, and the method comprises the steps: generating a real-time power utilization data sequence through collecting the electrical parameters of power utilization equipment of a pollution discharge enterprise in real time; performing statistical analysis on the real-time power consumption data sequence in the sliding window, and calculating the power consumption change rate, the power factor and the start-stop frequency in the corresponding window; constructing an equipment operation state recognition model by using a dynamic Bayesian network, and independently constructing a sub-model for each process stage to respectively perform parameter learning; inputting the multi-dimensional continuous features into a dynamic Bayesian network, and outputting the start-stop state of the equipment at the current moment; and dynamically adjusting a threshold value for judging the start-stop state of the equipment. The dynamic change of the start-stop state of the equipment is flexibly handled, the problem that a fixed threshold cannot adapt to the operation fluctuation of the equipment is avoided, and the judgment precision of the start-stop state of the equipment is ensured; the difficulty of cluster number selection is avoided, and the diversity and complexity of the operation state of the equipment can be better reflected.

Description

Technical Field

[0001] This application belongs to the technical field of equipment monitoring and status recognition, and more particularly relates to a method, device, and equipment for identifying the operating status of equipment in sewage enterprises. Background Art

[0002] With the continuous advancement of the industrialization process, the supervision of environmental protection efforts has gradually increased, and a series of strict environmental protection policies have been introduced, requiring enterprises to install pollution treatment equipment and discharge pollutants scientifically. To ensure the effective operation of production suspension and output restriction measures, power consumption monitoring technology has emerged. By analyzing the power consumption data of enterprises, it is possible to indirectly understand the operating conditions of production facilities and pollution treatment facilities, thereby quickly identifying and correcting potential sewage discharge violations, significantly improving the efficiency and accuracy of supervision.

[0003] Traditional methods for identifying the operating status of facilities mainly rely on preset power consumption thresholds for determination. Once the actual power consumption data reaches these threshold limits, the operating status of the facilities can be quickly determined. With the progress of monitoring technology, clustering algorithms have been widely applied in the field of power consumption analysis. As an unsupervised learning method, clustering algorithms deeply explore the internal characteristics of facility power consumption data and can automatically classify data with similar power consumption patterns into different operating status categories, thus achieving a more accurate and flexible judgment of the start-stop status of facilities.

[0004] Although the threshold judgment method is simple and intuitive, its core lies in presetting fixed power consumption thresholds as the judgment criteria. However, the operating status of facilities is not static but changes dynamically with various factors such as production demand and environmental conditions. Therefore, it is difficult for the threshold judgment method to flexibly adapt to this dynamic nature, and it may lead to the inability to make timely and accurate judgments when there is a deviation between the actual operating status of the facilities and the preset thresholds. In addition, for some special states, such as when the facilities are in standby or low-load operation states, their power consumption data may not be significant, making it difficult to effectively identify them through fixed thresholds. On the other hand, as a more advanced recognition algorithm, the clustering analysis method has shown certain advantages in dealing with complex and variable power consumption data, but it also faces problems such as insufficient result interpretability and difficulty in determining the number of clusters. The clustering results are often presented in a mathematical form, lacking intuitive interpretability and requiring in-depth interpretation in combination with practical experience and professional knowledge. In addition, how to determine the optimal number of clusters is also a challenge. Too many clusters may lead to information redundancy and difficult interpretation, while too few clusters may not accurately reflect the diversity of the operating status of facilities. Summary of the Invention

[0005] In view of the above analysis, the embodiments of the present invention aim to provide a method, device, and equipment for identifying the operating status of equipment in sewage enterprises, aiming to solve the challenges faced by the threshold judgment method and clustering analysis method in the prior art. For the dynamic changes in the start-stop status of equipment and the identification of special operating states, a more flexible and accurate solution is provided to improve the adaptability, accuracy, and operability of the system in complex environments.

[0006] In the first aspect of the present application, a method for identifying the operating status of equipment in sewage enterprises is provided, including the following steps: Collect the electrical parameters of the electrical equipment in sewage enterprises in real time to generate a real-time power consumption data sequence; Construct a sliding window, perform statistical analysis on the real-time power consumption data sequence within the sliding window, and calculate the power consumption change rate, power factor, and start-stop frequency within the corresponding window; Use a dynamic Bayesian network to construct an equipment operating status identification model. The nodes of the dynamic Bayesian network include the equipment start-stop status at the current moment, the current power consumption, the power consumption change rate, the power factor, the start-stop frequency, and the equipment start-stop status at the previous moment; Adopt the following parameter learning mechanism to train the conditional probability distribution of the dynamic Bayesian network: perform stage identification on the historical data of the equipment in sewage enterprises, and divide it into multiple different process stages; independently construct sub-models for each process stage, identify the behavioral characteristics of the equipment in each process stage, and perform parameter learning respectively; Input the multi-dimensional continuous features composed of the real-time power consumption data sequence, the power consumption change rate, the power factor, and the start-stop frequency into the dynamic Bayesian network, perform state inference based on the dynamic Bayesian network, and output the equipment start-stop status at the current moment; Among them, the method further includes dynamically adjusting the threshold for judging the equipment start-stop status according to the changes in the equipment operating characteristics and the identification results.

[0007] Optionally, the independently constructing sub-models for each process stage, identifying the behavioral characteristics of the equipment in each process stage, and performing parameter learning respectively includes: When independently constructing sub-models for each process stage, use the expectation maximization algorithm to estimate the conditional probability distribution between the equipment start-stop status and other features within the corresponding stage, and optimize the parameters in the way of expectation maximization.

[0008] Optionally, the independently constructing sub-models for each process stage, identifying the behavioral characteristics of the equipment in each process stage, and performing parameter learning respectively further includes: When performing parameter estimation, obtain labeled equipment power consumption sample data and unlabeled equipment power consumption sample data; Input the labeled device power consumption sample data and the unlabeled device power consumption sample data into the device operation status recognition model for semi-supervised learning training.

[0009] Optionally, the stage identification of the historical data of the pollution discharge enterprise equipment includes: Using a clustering algorithm to perform clustering analysis on the historical data to identify different operation stages of the equipment; the operation stages include a working stage, a standby stage, and a cleaning stage.

[0010] Optionally, the dynamic adjustment of the threshold for judging the start-stop state of the equipment according to the changes in the equipment operation characteristics and the recognition results includes: Based on the power consumption change rate , the threshold for judging the start-stop state of the equipment is dynamically updated using the following formula: ; where is the forgetting factor, is the threshold for judging the start-stop state of the equipment at the current moment, is the threshold for judging the start-stop state of the equipment at the previous moment, is the power consumption change rate at the current moment; Or real-time monitoring the false alarm rate and the missed alarm rate of the operation status of the pollution discharge enterprise equipment, and dynamically adjusting the threshold for judging the start-stop state of the equipment based on ; where is the threshold for judging the start-stop state of the equipment at the current moment, is the threshold for judging the start-stop state of the equipment at the previous moment, is the false alarm rate at the current moment, is the target false alarm rate; is the missed alarm rate at the current moment, is the target missed alarm rate, , is the adjustment coefficient; Or using a dynamic Bayesian network to calculate the posterior probability distribution of the equipment being in the enabled state , calculating the mean and standard deviation of the posterior probability distribution; dynamically adjusting the threshold for judging the start-stop state of the equipment in the following manner: ; where is the threshold for judging the start-stop state of the equipment at the current moment, is the posterior probability that the equipment is currently in the enabled state under the condition of observing the time series to , is the average value of the posterior probability, is the standard deviation of the posterior probability, is the confidence adjustment coefficient.

[0011] Optionally, it further includes: Adopt any of the following strategies to adaptively adjust the forgetting factor as follows: Adopt grid search within a preset range and combine it with cross-validation. By comparing the recognition accuracy, F1 score or false alarm rate metrics under different values, select the optimal value; Use the Bayesian optimization method to construct a surrogate model between and the performance metrics, and select the optimal value based on the historical recognition performance; According to the deviation between the current recognition accuracy and the target accuracy, dynamically adjust , and the update formula is: ; where is the learning rate, is the current recognition accuracy, is the target accuracy.

[0012] Optionally, it further includes: Based on the analysis of the real-time power consumption data sequence, judge the degree of fluctuation of the power consumption data; Dynamically adjust the length of the sliding window according to the degree of fluctuation of the power consumption data: calculate the degree of fluctuation of the power consumption data within the sliding window; when the degree of fluctuation of the power consumption data exceeds the set threshold, reduce the sliding window according to the preset step; when the degree of fluctuation of the power consumption data is lower than the set threshold, increase the sliding window according to the preset step.

[0013] In the second aspect of the present application, a device for identifying the operating state of equipment of a polluting enterprise is provided, including the following modules: A data acquisition module, configured to collect electrical parameters of the power consumption equipment of the polluting enterprise in real time and generate a real-time power consumption data sequence; A statistical calculation module, configured to construct a sliding window, perform statistical analysis on the real-time power consumption data sequence within the sliding window, and calculate the power consumption change rate, power factor and start-stop frequency within the corresponding window; A model construction module, configured to construct an equipment operating state recognition model by using a dynamic Bayesian network, and the nodes of the dynamic Bayesian network include the equipment start-stop state at the current moment, the current power consumption, the power consumption change rate, the power factor, the start-stop frequency and the equipment start-stop state at the previous moment; A parameter learning module, configured to train the conditional probability distribution of a dynamic Bayesian network by adopting the following parameter learning mechanism: perform stage identification on the historical data of the equipment of the polluting enterprise, and divide it into multiple different process stages; independently construct sub-models for each process stage, identify the behavior characteristics of the equipment in each process stage, and perform parameter learning respectively; An output module, configured to input the multi-dimensional continuous features composed of the real-time power consumption data sequence, the power consumption change rate, the power factor, and the start-stop frequency into the dynamic Bayesian network, perform state inference based on the dynamic Bayesian network, and output the start-stop state of the equipment at the current moment; Wherein, the device further includes: a threshold adjustment module, configured to dynamically adjust the threshold for judging the start-stop state of the equipment according to the changes in the equipment operation characteristics and the recognition results.

[0014] In the third aspect of the present application, there is provided a device for identifying the operation state of equipment of a polluting enterprise, including a memory and a processor, where the memory stores a computer program, and when the computer program is executed by the processor, it implements any one of the above-mentioned methods for identifying the operation state of equipment of a polluting enterprise.

[0015] In the fourth aspect of the present application, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the method for identifying the operation state of equipment of a polluting enterprise according to any one of the above.

[0016] The method for identifying the operation state of equipment of a polluting enterprise provided by the present application adopts a dynamic Bayesian network and a dynamic threshold adjustment mechanism, and can automatically adjust the threshold in real time according to the changes in the equipment operation characteristics and the recognition results. In this way, it can flexibly cope with the dynamic changes in the start-stop state of the equipment, avoid the problem that a fixed threshold cannot adapt to the operation fluctuations of the equipment, and ensure the accuracy and flexibility of the judgment of the start-stop state of the equipment. By combining features such as the power consumption change rate, the threshold can be sensitively adjusted when the equipment is in a standby state, a low-load state or other special states, ensuring that these low-fluctuation states can also be accurately identified, thereby improving the comprehensiveness of the identification.

[0017] By introducing a dynamic Bayesian network and a real-time data feedback mechanism, this application can provide a more intuitive and operable decision-making basis for identifying the start-stop state of equipment. The application of the dynamic threshold adjustment mechanism enables the identification of the start-stop state of equipment to not only rely on complex clustering results but also combine the electricity consumption data and changes of actual equipment, providing a judgment criterion that is easier to understand and apply. Different from the traditional clustering analysis method, this application optimizes the identification strategy of the start-stop state of equipment through an adaptive approach, combining real-time data feedback and a dynamic threshold adjustment mechanism, rather than simply relying on a preset number of clusters. In this way, it is possible to flexibly adjust the judgment strategy according to the actual operating state of the equipment, avoiding the difficulty of choosing the number of clusters and being able to better reflect the diversity and complexity of the equipment operating state.

[0018] In addition, this application also provides a device and equipment for identifying the operating state of equipment in sewage disposal enterprises with the above technical effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the embodiments of this specification. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a flowchart of a specific implementation manner of the method for identifying the operating state of equipment in sewage disposal enterprises provided by this application; Figure 2 It is a flowchart of another specific implementation manner of the method for identifying the operating state of equipment in sewage disposal enterprises provided by this application; Figure 3 It is a schematic diagram of the electricity consumption data sequence and operating state identification of the production facilities of an example enterprise; Figure 4 It is a structural block diagram of the device for identifying the operating state of equipment in sewage disposal enterprises provided by this application; Figure 5 It is a structural block diagram of the equipment for identifying the operating state of equipment in sewage disposal enterprises provided by this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, rather than all of them. It should be noted that, without conflict, the implementation manners and features in the implementation manners in this disclosure can be combined, separated, interchanged, and / or rearranged. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0022] The terms used herein are for the purpose of describing specific embodiments and are not intended to be restrictive. As used herein, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are also intended to include the plural forms. In addition, when the terms "comprise" and / or "include" and their variants are used in this specification, it is stated that there are the stated features, wholes, steps, operations, components, assemblies, and / or groups thereof, but it does not exclude the presence or addition of one or more other features, wholes, steps, operations, components, assemblies, and / or groups thereof. It should also be noted that, as used herein, the terms "substantially", "about", and other similar terms are used as approximate terms rather than degree terms, so they are used to explain the inherent deviations of measured values, calculated values, and / or provided values that those of ordinary skill in the art will recognize.

[0023] This embodiment provides a method for identifying the operating status of equipment in sewage-discharging enterprises, which predicts the start-stop status of equipment and adaptively adjusts thresholds based on real-time electricity consumption data, electricity consumption change rate, and a dynamic Bayesian network (DBN) model. Refer to Figure 1 , and the process specifically includes the following steps: S101: Real-time collect the electrical parameters of the electricity-consuming equipment of sewage-discharging enterprises to generate a real-time electricity consumption data sequence.

[0024] In this embodiment, first, the intelligent electricity meters installed on the key electricity-consuming equipment of sewage-discharging enterprises are used to real-time collect the electrical parameters of the equipment, including but not limited to current, voltage, power, etc. These data are real-time transmitted to the data collection system through Internet of Things (IoT) devices to generate a real-time electricity consumption data sequence, providing data support for subsequent analysis.

[0025] After the real-time collection of the electrical parameters of the electricity-consuming equipment of sewage-discharging enterprises, it further includes: cleaning the original data of the electrical parameters, removing null values, duplicate data, and outliers to obtain the cleaned data for subsequent data processing.

[0026] S102: Construct a sliding window, perform statistical analysis on the real-time power consumption data sequence within the sliding window, and calculate the power consumption change rate, power factor, and start-stop frequency within the corresponding window.

[0027] After data acquisition, the real-time power consumption data sequence is processed by the sliding window analysis method. The length of the sliding window can be adjusted according to actual needs and is usually in the range of dozens of seconds to several minutes. Perform statistical analysis on the power consumption data within the sliding window to calculate the power consumption change rate ( ). The power consumption change rate reflects the fluctuation of the equipment's power consumption, and the calculation formula is: where, is the power consumption at the current moment, is the power consumption at the previous moment. The magnitude of the power consumption change rate can reveal whether the equipment is in a period of start-stop state change or large load fluctuation.

[0028] S103: Use a dynamic Bayesian network to construct a device operation state recognition model. The nodes of the dynamic Bayesian network include the device start-stop state at the current moment, the current power consumption, the power consumption change rate, the power factor, the start-stop frequency, and the device start-stop state at the previous moment. Using a dynamic Bayesian network to construct a device operation state recognition model includes: determining the nodes in the dynamic Bayesian network, and each node represents a variable. Specifically, the following nodes are included: Device current state node ( ): This node represents the start-stop state of the device at the current moment, and the value of 1 represents start, and 0 represents stop. The device start-stop state is the target variable to be predicted.

[0029] Device power consumption data node ( ): This node represents the power consumption of the device at the current moment and is used as an observed variable to reflect the operation intensity and load change of the device.

[0030] Power consumption change rate node ( ): This node represents the power consumption change rate of the device at the current moment and is used to capture the fluctuation of the load change during the device operation to help infer whether the device is in the process of start-stop state change.

[0031] Threshold node ( ): This node represents the threshold for judging the device start-stop state, depends on the device's power consumption data and change rate, and will be dynamically adjusted as the device operation state changes.

[0032] Construct a directed graph model of the dynamic Bayesian network, including the following two types of dependency relationships: Dependency relationship within a moment: The current start-stop state of the device ( ) has a direct dependence on the current electricity consumption ( ) and the electricity consumption change rate ( ). That is, the start-stop state of the device is not only related to the current electricity consumption, but also closely related to the change of electricity consumption (change rate). Specifically, when the device is operating at high load, it may be in the start state, and when the load of the device changes (such as a sudden drop), it may lead to a state switch (such as shutdown or standby).

[0033] In the directed graph of the DBN, the current state node ( ) directly points to the electricity consumption data node ( ) and the electricity consumption change rate node ( ), forming a dependence relationship within a moment, indicating that the start-stop state of the device at the current moment is jointly affected by the electricity consumption and the electricity consumption change rate.

[0034] Cross-time dependence relationship: The current start-stop state of the device ( ) not only depends on the current electricity consumption data ( ) and the electricity consumption change rate ( ), but also has a dependence relationship with the state node at the previous moment ( ). The current state of the device is usually affected by its historical state. Especially during the operation of the device, the start-stop state of the device at the previous moment may affect the transition of the current state.

[0035] After the network structure is determined, by inputting real-time electricity consumption data ( ) and the electricity consumption change rate ( ) into the dynamic Bayesian network model, the start-stop state of the device ( ) is inferred. During the inference process, the DBN model uses the within-moment dependence relationship and the cross-time dependence relationship, combines the historical state of the device ( ) and the current electricity consumption data and change rate, and predicts the start-stop state of the device at the current moment.

[0036] Through the forward-backward inference algorithm or the Bayesian inference method, the current state of the device (S t ) can be inferred based on the historical data and the real-time data, and the prediction result of the start-stop state of the device is continuously updated.

[0037] S104: Use the following parameter learning mechanism to train the conditional probability distribution of the dynamic Bayesian network: Identify the historical data of the equipment of the polluting enterprise by stage, and divide it into multiple different process stages; Independently construct sub-models for each process stage, identify the behavioral characteristics of the equipment in each process stage, and perform parameter learning respectively.

[0038] Specifically, a clustering algorithm can be used to perform clustering analysis on the historical data to identify different operating stages of the device; the operating stages include the working stage, the standby stage, and the cleaning stage. This process can ensure that the behaviors within each stage are independently modeled. In this process, a clustering algorithm (such as K-means or DBSCAN) is used to perform clustering analysis on the historical data of the device to identify the behavioral characteristics of the device in different process stages. For example, based on features such as the start-stop mode, power consumption, and power factor of the device, different process segments such as working, cleaning, and standby are identified. To improve accuracy, a time series model (such as Hidden Markov Model, HMM) can also be used to classify the device state, so as to better capture the periodic changes of the device.

[0039] When independently constructing sub-models for each process stage, the Expectation-Maximization (EM) algorithm is used to estimate the conditional probability distribution between the start-stop state of the device and other features within the corresponding stage, and the parameters are optimized in an Expectation-Maximization manner.

[0040] Once the data is divided into different process segments, parameter learning can be independently performed within each stage. For example, in the working stage, the device may require higher power to maintain the working state, while in the cleaning stage, the power factor and power consumption may decrease significantly. The parameters of the Bayesian network can be trained separately within each stage, so that the models for each process segment can better adapt to specific device behaviors.

[0041] When independently performing parameter learning within each stage, the EM algorithm can help determine the conditional probability distribution at each moment. Specifically, assuming that the state of the device (such as the start-stop state) has different behavioral patterns in different stages, the EM algorithm can be used to estimate the conditional probability distribution within each stage. For example, for the working stage, the model can estimate the conditional probability distribution of power consumption under the current start-stop state; while in the cleaning stage, the model can learn different patterns.

[0042] When performing parameter estimation, labeled device power consumption sample data and unlabeled device power consumption sample data are obtained; the labeled device power consumption sample data and the unlabeled device power consumption sample data are input into the device operating state recognition model for semi-supervised learning training. When performing parameter learning, by combining semi-supervised learning methods, the model performance can be further improved through pseudo-labels and unlabeled data, especially in the case of insufficient data annotation.

[0043] S105: Input the multi-dimensional continuous features composed of the real-time power consumption data sequence, the power consumption change rate, the power factor, and the start-stop frequency into the dynamic Bayesian network, perform state inference based on the dynamic Bayesian network, and output the start-stop state of the device at the current moment.

[0044] The multi-dimensional continuous features composed of the power consumption data sequence collected in real time, the calculated power consumption change rate ( ), power factor, and start-stop frequency are input into the dynamic Bayesian network, and the current state of the device is predicted according to the inference algorithm (such as the forward-backward algorithm) ( ). During the inference process, combining historical data and real-time data, the start-stop state (start or stop) of the device is output for each time point.

[0045] In the embodiments of the present application, the start-stop frequency (indicating the number of state switches per unit time) and the power factor change (reflecting whether the device is actually loaded) are introduced into the input features. These features help to identify behaviors such as idling operation. The constructed DBN nodes further include: a start-stop frequency node and a power factor node.

[0046] After receiving the input features, the following modeling method is adopted in the DBN: The conditional probability of each continuous variable is represented using a conditional Gaussian distribution; or the joint probability distribution of the states under multi-dimensional input is represented using a Gaussian mixture model; the learning process maximizes the likelihood through the EM algorithm to train the mean, variance, and covariance matrix of the input variables in each state.

[0047] In this embodiment, by constructing multi-dimensional continuous features (power consumption change rate, power factor, start-stop frequency) as the input observation variables of the dynamic Bayesian network, and based on the within-time and cross-time dependence relationships, combined with historical states for state inference, the start-stop state of the device can be identified more accurately and dynamically. This method has strong expressive power, sensitivity, and adaptability, significantly superior to the traditional single-threshold start-stop judgment method. Among them, the method further includes: dynamically adjusting the threshold for judging the start-stop state of the device according to the changes in the device operation characteristics and the recognition results.

[0048] During the inference of the start-stop state of the device, by calculating the power consumption change rate of the device in real time, and according to the fluctuation of the power consumption change rate, the threshold for judging the start-stop state ( )is adjusted in real time: judge whether the change rate of the device power consumption data exceeds a predetermined threshold within a preset time interval. If so, increase the size of the threshold according to a preset step; if the change rate of the device power consumption data does not detect a change in the start-stop state within the preset time interval, reduce the size of the threshold according to a preset step.

[0049] When the power consumption change rate is large (i.e., the start-stop state of the device changes frequently), increase the threshold ( ), so as to reduce the sensitivity to small fluctuations and avoid false alarms. When the power consumption change rate is small (i.e., the device runs smoothly for a long time), reduce the threshold ( ), to improve the sensitivity to small changes and ensure timely capture of potential start-stop changes of the device.

[0050] Specifically, any one or any combination of the following strategies can be adopted to dynamically adjust the threshold for judging the start-stop state of the device: The first strategy is dynamic adjustment based on the electricity consumption change rate. Specifically: based on the electricity consumption change rate , the following formula is used to dynamically update the threshold for judging the start-stop state of the device: .

[0051] Among them, is the forgetting factor, is the threshold for judging the start-stop state of the device at the current moment, is the threshold for judging the start-stop state of the device at the previous moment, is the electricity consumption change rate at the current moment.

[0052] It can be understood that is used to control the weight between historical memory and current feedback. If the value is small, it can quickly respond to the current change; If the value is large, it is more dependent on history and the response is smoother. Further, any of the following strategies can be adopted to adaptively adjust the forgetting factor : (1) Adopt grid search and combine cross-validation within a preset range (such as 0.1~0.9), and select the optimal value by comparing the recognition accuracy, F1 score or false alarm rate indicators under different values.

[0053] (2) Use the Bayesian optimization method (such as Gaussian Process) to construct a surrogate model between and the performance index, and select the optimal value based on the historical recognition performance. This method can improve the search efficiency; the advantage is that it is applicable to high-dimensional parameter spaces and non-convex objective functions.

[0054] (3) According to the deviation between the current recognition accuracy and the target accuracy, dynamically adjust , and the update formula is: ; among them, is the learning rate, is the current recognition accuracy, is the target accuracy.

[0055] In a preferred embodiment of the present invention, a linkage control mechanism based on performance feedback is further constructed to simultaneously dynamically adjust the threshold for judging the start-stop state of the device ( ) and the forgetting factor ( ), to enhance the adaptive ability to operating fluctuations and error changes. Specifically, based on the deviation relationship between the current recognition result and the target performance indicators (such as false alarm rate, miss rate, accuracy), the threshold and value are dynamically updated.

[0056] The strategy is as follows: When it is detected that the current false alarm rate is higher than the set target false alarm rate, it means that the judgment is too sensitive and prone to false activation judgments. Therefore, the control strategy is: increase the judgment threshold , and decrease the forgetting factor , so as to respond to new changes more quickly and reduce false alarms. When the current miss rate is higher than the target miss rate, it means that the judgment is too conservative and fails to identify the activation status of the device in a timely manner. At this time, the judgment threshold will be decreased, and will also be decreased, enhancing the sensitivity to fluctuating signals. When the current accuracy is stably higher than the set target accuracy, it indicates that the operation is stable and the recognition effect is ideal. To avoid overresponding to short-term fluctuations, is appropriately increased at this time to enhance the smoothness and stability of the system.

[0057] The second strategy is to dynamically adjust by real-time monitoring of the false alarm rate and miss rate of the system. In the case of an increasing false alarm rate, increase the threshold and reduce the system sensitivity; in the case of an increasing miss rate, decrease the threshold and increase the system sensitivity. Specifically: Real-time monitor the false alarm rate and miss rate of the operation status of the equipment of polluting enterprises, and dynamically adjust the threshold for judging the start-stop status of the equipment based on .

[0058] Among them, is the threshold for judging the start-stop status of the equipment at the current moment, is the threshold for judging the start-stop status of the equipment at the previous moment, is the false alarm rate at the current moment, is the target false alarm rate; is the miss rate at the current moment, is the target miss rate, , is the adjustment coefficient.

[0059] The third is to dynamically adjust the threshold based on the posterior probability distribution inferred by DBN. If the confidence distribution is too concentrated, indicating that the model is confident, the threshold can be increased; if the distribution is more dispersed, indicating that the model is uncertain, the threshold should be decreased to increase the sensitivity. Specifically: Use the dynamic Bayesian network to calculate the posterior probability distribution of the equipment being in the activated state , calculate the mean and standard deviation of the posterior probability distribution; dynamically adjust the threshold for judging the start-stop status of the equipment in the following way: .

[0060] Among them, is the threshold for judging the start / stop state of the device at the current moment, is for time series observation up to Under the condition of, the posterior probability that the device is currently in the enabled state, is the average value of the posterior probability, is the standard deviation of the posterior probability, is the confidence adjustment coefficient, which can be used to control conservatism: the larger γ is, the more the threshold tends to tolerate errors.

[0061] The threshold adjustment is optimized through a dynamic Bayesian network, which can dynamically adjust the judgment criteria according to the actual operating state of the device to ensure the accuracy of state recognition.

[0062] As the device continues to operate, the system continuously adjusts and optimizes the threshold through a real-time feedback mechanism to adapt to the long-term changes of the device. For example, during the operation of the device, if the system finds that some devices start and stop frequently or run smoothly for a long time, it will adjust the update rate and range of the threshold through real-time analysis of historical data to further improve the detection sensitivity and accuracy of the device start / stop state.

[0063] Through the above implementation steps, this embodiment provides a method for identifying the operating state of sewage enterprise equipment based on a dynamic Bayesian network and a dynamic threshold adjustment mechanism. This method can flexibly respond to the dynamic changes of equipment operation, accurately identify the start / stop state of the equipment, and continuously optimize the threshold through a real-time feedback mechanism, thereby improving the efficiency and accuracy of the equipment monitoring system.

[0064] The flowchart of another specific implementation manner of the method for identifying the operating state of sewage enterprise equipment provided by this application is as Figure 2 shown, and this method specifically includes: S201: Collect raw data.

[0065] Install smart meters on the key power-consuming equipment of the enterprise to capture multiple parameters such as current, voltage, power, and electricity in real time.

[0066] S202: Preprocess the raw data.

[0067] Clean the raw data collected in real time to remove null values, duplicate data, and obvious outliers (such as sudden spike data caused by sensor failures or interference).

[0068] S203: Calculate the instantaneous change rate.

[0069] Based on historical electricity consumption data, calculate the change rate of each time segment.

[0070] The calculation method of the instantaneous change rate is as follows: Among them, among them, is the power value at time t, is the time difference between two time points.

[0071] S204: Perform trend analysis and start / stop state detection.

[0072] (1) Trend analysis By dynamically adjusting the size of the sliding window, adjusting the window length according to the severity of data fluctuations, and using regression to judge the trend within the window.

[0073] Specifically, based on the analysis of the real-time electricity consumption data sequence, judge the degree of fluctuation of the electricity consumption data; dynamically adjust the length of the sliding window according to the degree of fluctuation of the electricity consumption data: calculate the degree of fluctuation of the electricity consumption data within the sliding window; when the degree of fluctuation of the electricity consumption data exceeds the set threshold, reduce the sliding window according to the preset step size; when the degree of fluctuation of the electricity consumption data is lower than the set threshold, increase the sliding window according to the preset step size. The specific implementation of this process can be: 1) Initialize the window: Define the initial sliding window size W min and W max which represent the minimum and maximum window sizes respectively.

[0074] 2) Calculate the fluctuation characteristics within the window: For each sliding window, calculate the degree of data fluctuation within the window, such as the standard deviation or variance. The larger the standard deviation, the more severe the fluctuation; the smaller the standard deviation, the more stable the data.

[0075] 3) Dynamically adjust the window size: Set a fluctuation threshold. When the standard deviation exceeds the threshold, reduce the window; when the standard deviation is lower than the threshold, expand the window.

[0076] 4) For each window, linear regression or polynomial regression can be used to fit the electricity consumption trend line to observe whether the trend shows an upward, downward or stable state.

[0077] (2) Start / stop state detection By setting an initial threshold (which can be determined according to the device historical data) to judge the start / stop state of the device. When the change rate exceeds this threshold, it is considered that the device starts / stops.

[0078] S205: Dynamically adjust the threshold.

[0079] As data accumulates, the threshold for change detection is adjusted in real time through algorithms such as EWMA, machine learning or reinforcement learning.

[0080] In the following embodiments, taking the Dynamic Bayesian Network (DBN) as an example, the training process of the model is as follows: (1) Data Preparation Collect historical data of the device, including time-series power consumption data, start-stop status, and other relevant features. Extract features from the original data, such as the power consumption change rate, standard deviation, variance, etc. at each moment, and label the data. (2) Constructing a Dynamic Bayesian Network DBN is a directed graph model that can represent time-dependent relationships. Its structure is divided into dependencies between moments (cross-time dependencies) and dependencies within moments (same-time dependencies). When modeling the power consumption of the device, the basic structure of DBN can be expressed as: Device start-stop status: Represents the start-stop status of the device (S t ), which is the variable to be predicted.

[0081] Device power consumption: Represents the power consumption at each time point (P t ), which is the observed variable.

[0082] Change rate: The change rate of the device's power consumption (ΔP t ), which is a feature extracted from historical data.

[0083] Threshold adjustment: Represents the threshold (θ t ), which is the parameter to be dynamically adjusted for judging the start-stop status of the device.

[0084] 1) Defining DBN nodes Current state node ( ): The start-stop status of the device at the current time step, and the value can be start or stop (0 or 1).

[0085] Previous state node ( ): The start-stop status at the previous time step.

[0086] Current power consumption ( ): The power consumption of the device at the current time step.

[0087] Change rate node ( ): The change rate of the device's power consumption at the current time step.

[0088] Threshold node ( ): The change threshold for judging the start-stop status of the device, which depends on the historical data of the device.

[0089] 2) DBN modeling structure Assume that at time t, the state of the device is determined by the previous state and the current power consumption and the rate of change decide, and the rate-of-change threshold of the device is adaptively adjusted over time.

[0090] 3) Parameter learning The parameter learning of the DBN includes two parts: structure learning and parameter learning.

[0091] Structure learning: Use historical data to determine the dependency relationship between the start-stop state of the device and its features through algorithms (such as the Expectation-Maximization algorithm, EM).

[0092] Parameter learning: Determine the parameters of each conditional probability distribution in the Bayesian network according to historical data. The conditional probability distribution of S t can be expressed as , and use historical data to learn the parameters of these conditional probability tables through maximum likelihood estimation or Bayesian estimation.

[0093] 4) Inference and automatic threshold adjustment Infer the state of the device through the DBN model , and adjust the threshold according to historical data .

[0094] Infer the start-stop state of the device: Predict whether the device starts or stops at the current moment through the inference of the DBN model (such as the forward-backward algorithm) .

[0095] Adjust the threshold: During the inference process, automatically adjust the threshold using dynamic adjustment rules . The adjustment strategy can be based on the results of Bayesian inference and the feedback of historical data: During the process of inferring the start-stop state of the device, update the threshold for judging the start-stop state ( ) and historical data in real time according to the electricity consumption rate of change ( ) to optimize the sensitivity of identifying the start-stop state of the device. The dynamic threshold update follows the following rules: The dynamic threshold update formula is: .

[0096] Among them, is the forgetting factor, is the threshold used to judge the start-stop state of the device at the current moment, is the threshold used to judge the start-stop state of the device at the previous moment, is the electricity consumption rate of change at the current moment.

[0097] Forgetting factor ( ): Controls the influence degree of historical data on the current threshold adjustment. The value ranges from 0 to 1. When is close to 1, the influence of historical data is greater, and the threshold adjustment is strongly affected by the historical state; when is close to 0, the influence of the current power consumption change rate is greater, and the threshold adjustment is more dependent on real-time data.

[0098] Threshold adjustment mechanism: When the start-stop state of the device changes frequently, the power consumption change rate of the device ( ) is large, then the threshold ( ) is increased to reduce the response to small fluctuations; while when the start-stop state of the device is stable, the power consumption change rate is small, then the threshold ( ) is decreased to improve the sensitivity to small changes and ensure timely identification of potential changes in the start-stop state of the device.

[0099] By introducing a forgetting factor ( ) and a dynamic threshold update formula, the system can adaptively adjust the threshold according to real-time data, avoiding the limitation of the traditional fixed-threshold method that cannot adapt to changes in the start-stop state of the device.

[0100] 5) Model evaluation and optimization Using the historical data of the device, compare the difference between the start-stop state of the device predicted by the DBN model and the actual state, and calculate indicators such as accuracy, recall rate, and false alarm rate. According to different device characteristics, different threshold adjustment strategies can be selected through cross-validation to optimize the parameters .

[0101] Through historical data for model evaluation and optimization, including evaluating the difference between the predicted start-stop state of the device and the actual state, and optimizing the forgetting factor through cross-validation, finally select the best value of the forgetting factor to improve the performance of the model.

[0102] Compare the start-stop state of the device predicted by the model with the actual start-stop state in the historical data, and calculate the evaluation indicators. These evaluation indicators include: Accuracy: The proportion of the correctly predicted start-stop states by the model among all predicted states.

[0103] Recall rate: The proportion of the start-stop states predicted by the model and actually being start-stop states, measuring the ability of the model to identify changes in the start-stop state of the device.

[0104] False alarm rate: The proportion of the states wrongly predicted as start-stop states by the model, measuring the degree of false alarms of the model.

[0105] By calculating these evaluation indicators, the performance of the current DBN model in identifying the start-stop state of the device can be evaluated, and a basis for the next optimization can be provided.

[0106] After the evaluation of the device start-stop state prediction model, the forgetting factor is optimized by the cross-validation method to improve the performance of the model.

[0107] Cross-validation method: The historical data of the device is divided into k subsets by k-fold cross-validation. Each time, one subset is selected as the validation set, and the remaining subsets are used as the training set. Training and validation are repeated until each subset has been validated once. Cross-validation can avoid overfitting of the model and provide a more accurate performance evaluation.

[0108] Forgetting factor optimization: During the cross-validation process, different forgetting factor values are tested. The forgetting factor controls the influence degree of historical data on the current threshold adjustment, and its range is between 0 and 1. When is close to 1, the threshold update of the model depends more on historical data; when is close to 0, the threshold update depends more on the current real-time data.

[0109] Optimization strategy: In each round of cross-validation, calculate the evaluation metrics (such as accuracy, recall rate, and false alarm rate) under the current value.

[0110] By comparing the evaluation results under different values, select the best forgetting factor value to make the model perform optimally in the recognition of the device start-stop state. After the optimization process, output the best value and use it to update the threshold update mechanism in the dynamic Bayesian network model, so as to improve the adaptability of the model to the changes in the device start-stop state.

[0111] After the real-time output of the detected device start-stop state, it can further include the step of updating the log. Figure 3 Shows a schematic diagram of the electricity consumption data sequence and operation state recognition of an example enterprise production facility.

[0112] Under steady conditions, the statistical characteristics (such as the mean) of the facility's electricity consumption data remain unchanged over the time dimension, with only random fluctuations. However, when the facility starts or shuts down, the electricity consumption data sequence usually undergoes a structural break or change. These changes introduce non-stationarity, thus altering the overall distribution of the time series. The time points marking the beginning of these changes are called change points. The greatest advantage of the Dynamic Bayesian Network (DBN) lies in its ability to capture changes in data over the time dimension and continuously correct the model based on historical data. Through the inference of the hidden state (start-stop state), DBN adjusts the change rate threshold in real time to adapt to the actual operating conditions of the device, and dynamically adjusts the sliding window size according to the degree of data fluctuation. This mechanism dynamically changes the sliding window length by analyzing statistical features such as the standard deviation or variance, enabling the system to respond quickly when significant changes occur in the start-stop state of the device, and reducing over-sensitivity when the start-stop state of the device is stable.

[0113] By combining a sliding window, trend analysis, and a dynamic Bayesian network, this application can more accurately identify the start-stop state of the device. Especially during the start-stop process of the device, DBN provides higher detection accuracy and reduces misjudgments caused by short-term fluctuations. At the same time, the dynamic threshold adjustment mechanism automatically adjusts the threshold according to the real-time operating state of the device to ensure the optimization of detection sensitivity, further reducing the error probability and enabling better adaptation to the complex operating environments of different enterprises and different devices. This algorithm not only greatly improves the real-time performance, accuracy, and intelligent level of identifying the operating state of the facility, but also improves the supervision efficiency of abnormal behaviors of sewage-discharging enterprises, promoting the continuous progress of the environmental protection cause.

[0114] In some embodiments, the structural block diagram of the device for identifying the operating state of the sewage-discharging enterprise equipment provided by this application is as Figure 4 shown, and the device includes the following modules: A data acquisition module 401, configured to collect the electrical parameters of the electricity-consuming equipment of the sewage-discharging enterprise in real time and generate a real-time electricity consumption data sequence.

[0115] A statistical calculation module 402, configured to construct a sliding window, perform statistical analysis on the real-time electricity consumption data sequence within the sliding window, and calculate the electricity consumption change rate, power factor, and start-stop frequency within the corresponding window.

[0116] A model construction module 403, configured to construct a device operating state recognition model using a dynamic Bayesian network. The nodes of the dynamic Bayesian network include the start-stop state of the device at the current moment, the current electricity consumption, the electricity consumption change rate, the power factor, the start-stop frequency, and the start-stop state of the device at the previous moment.

[0117] The parameter learning module 404 is configured to train the conditional probability distribution of the dynamic Bayesian network by adopting the following parameter learning mechanism: perform stage identification on the historical data of the equipment of the polluting enterprise, and divide it into multiple different process stages; independently construct sub-models for each process stage, identify the behavior characteristics of the equipment in each process stage, and perform parameter learning respectively.

[0118] The output module 405 is configured to input the multi-dimensional continuous features composed of the real-time power consumption data sequence, the power consumption change rate, the power factor, and the start-stop frequency into the dynamic Bayesian network, perform state inference based on the dynamic Bayesian network, and output the start-stop state of the equipment at the current moment.

[0119] Wherein, the device further includes: a threshold adjustment module, configured to dynamically adjust the threshold for judging the start-stop state of the equipment according to the changes in the equipment operation characteristics and the recognition result.

[0120] An embodiment of the present application further provides a device for identifying the operation state of the equipment of a polluting enterprise, as Figure 5 shown. The device includes a memory 51 and a processor 52. The memory 51 stores a computer program, and when the computer program is executed by the processor 52, it realizes the method for identifying the operation state of the equipment of a polluting enterprise according to any one of the above.

[0121] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the method for identifying the operation state of the equipment of a polluting enterprise according to any one of the above.

[0122] Computer-readable storage media include both permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology for storing information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0123] Those skilled in the art should also be able to further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0124] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0125] The specific implementation manners described above have further elaborated on the purpose, technical solution, and beneficial effects of this application. It should be understood that the above is only the specific implementation manner of this application and is not used to limit the protection scope of this application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this application shall be included in the protection scope of this application.

Claims

1. A method for identifying the operating state of equipment in sewage enterprises, characterized in that, It includes the following steps: Collect the electrical parameters of the power-consuming equipment of the polluting enterprises in real time to generate a real-time power consumption data sequence; Construct a sliding window, perform statistical analysis on the real-time power consumption data sequence within the sliding window, and calculate the power consumption change rate, power factor, and start-stop frequency within the corresponding window; Construct a device operation state recognition model using a dynamic Bayesian network. The nodes of the dynamic Bayesian network include the device start-stop state at the current moment, the current power consumption, the power consumption change rate, the power factor, the start-stop frequency, and the device start-stop state at the previous moment; Adopt the following parameter learning mechanism to train the conditional probability distribution of the dynamic Bayesian network: identify different process stages for the historical data of the polluting enterprise equipment, and divide them into multiple different process stages; independently construct sub-models for each process stage, identify the behavioral characteristics of the equipment in each process stage, and perform parameter learning separately; Input the multi-dimensional continuous features composed of the real-time power consumption data sequence, the power consumption change rate, the power factor, and the start-stop frequency into the dynamic Bayesian network, perform state inference based on the dynamic Bayesian network, and output the device start-stop state at the current moment; Among them, the method further includes dynamically adjusting the threshold for judging the device start-stop state according to the changes in the device operation characteristics and the recognition results.

2. The method for identifying the operation state of the equipment of a sewage discharging enterprise according to claim 1, characterized in that, The independently constructing sub-models for each process stage, identifying the behavioral characteristics of the equipment in each process stage, and performing parameter learning separately includes: When independently constructing sub-models for each process stage, use the expectation maximization algorithm to estimate the conditional probability distribution between the device start-stop state and other features within the corresponding stage, and optimize the parameters in the way of expectation maximization.

3. The method for identifying the operating state of the equipment of a sewage discharge enterprise according to claim 2, characterized in that, The independently constructing sub-models for each process stage, identifying the behavioral characteristics of the equipment in each process stage, and performing parameter learning separately further includes: When performing parameter estimation, obtain the labeled device power consumption sample data and the unlabeled device power consumption sample data; Input the labeled device power consumption sample data and the unlabeled device power consumption sample data into the device operation state recognition model for semi-supervised learning training.

4. The method for identifying the operation state of the equipment of a sewage discharge enterprise according to claim 2, wherein, The identifying different process stages for the historical data of the polluting enterprise equipment includes: Adopt a clustering algorithm to perform clustering analysis on the historical data to identify different operation stages of the equipment; the operation stages include the working stage, the standby stage, and the cleaning stage.

5. The method for identifying the operation state of the equipment of a sewage discharge enterprise according to any one of claims 1 to 4, characterized in that, The dynamically adjusting the threshold for judging the device start-stop state according to the changes in the device operation characteristics and the recognition results includes: Based on the electricity consumption change rate , the threshold value used to judge the start-stop state of the device is dynamically updated by using the following formula: ; Among them, is the forgetting factor, is the threshold value for judging the start-stop state of the device at the current moment, is the threshold value for judging the start-stop state of the device at the previous moment, is the power consumption change rate at the current moment; Or the false alarm rate and missed alarm rate of real-time monitoring of the operation status of pollution discharge enterprises' equipment, based on dynamically adjust the threshold for judging the start-stop status of the equipment; Among them, is the threshold value for judging the start / stop state of the device at the current moment, is the threshold value for judging the start / stop state of the device at the previous moment, is the false alarm rate at the current moment, is the target false alarm rate; is the missed alarm rate at the current moment, is the target missed alarm rate, , is the adjustment coefficient; Or calculate the posterior probability distribution of the device being in the enabled state using a dynamic Bayesian network, and calculate the mean and standard deviation of the posterior probability distribution; dynamically adjust the threshold for judging the start / stop state of the device in the following manner: The posterior probability distribution, calculate the mean and standard deviation of the posterior probability distribution; dynamically adjust the threshold for judging the start / stop state of the device in the following manner: ； Among them, is the threshold for judging the start / stop state of the device at the current moment, is the posterior probability that the device is currently in the enabled state under the condition of time series observation to , is the average value of the posterior probability, is the standard deviation of the posterior probability, is the confidence adjustment coefficient.

6. The method for identifying the operating state of the equipment of a sewage discharge enterprise according to claim 5, characterized in that, It also includes: Adopt any of the following strategies for the forgetting factor to perform adaptive adjustment: Within a preset range, grid search is adopted in combination with cross-validation. By comparing the recognition accuracy, F1 score, or false alarm rate metrics at different values, the optimal value is selected; Using the Bayesian optimization method, a surrogate model is constructed between and the performance metrics, and the optimal value is selected based on the historical recognition performance; Dynamically adjust according to the deviation between the current recognition accuracy and the target accuracy , and the update formula is: ; wherein, is the learning rate, is the current recognition accuracy rate, is the target accuracy rate.

7. The method for identifying the operation state of the equipment of a sewage discharge enterprise according to claim 5, characterized in that, It also includes: Based on the analysis of the real-time power consumption data sequence, judge the fluctuation degree of the power consumption data; Dynamically adjust the length of the sliding window according to the fluctuation degree of the power consumption data: calculate the fluctuation degree of the power consumption data within the sliding window; when the fluctuation degree of the power consumption data exceeds the set threshold, reduce the sliding window according to the preset step; when the fluctuation degree of the power consumption data is lower than the set threshold, increase the sliding window according to the preset step.

8. An identification device for the operation state of equipment of a sewage discharge enterprise, characterized in that, It includes the following modules: A data acquisition module configured to collect the electrical parameters of the power-consuming equipment of the polluting enterprises in real time to generate a real-time power consumption data sequence; A statistical calculation module, configured to construct a sliding window, perform statistical analysis on the real-time power consumption data sequence within the sliding window, and calculate the power consumption change rate, power factor, and start-stop frequency within the corresponding window; A model construction module, configured to construct a device operation state recognition model using a dynamic Bayesian network, where the nodes of the dynamic Bayesian network include the device start-stop state at the current moment, the current power consumption, the power consumption change rate, the power factor, the start-stop frequency, and the device start-stop state at the previous moment; A parameter learning module, configured to train the conditional probability distribution of the dynamic Bayesian network using the following parameter learning mechanism: perform stage identification on the historical data of the sewage discharge enterprise's equipment, and divide it into multiple different process stages; independently construct sub-models for each process stage, identify the behavior characteristics of the equipment in each process stage, and perform parameter learning respectively; An output module, configured to input the multi-dimensional continuous features composed of the real-time power consumption data sequence, the power consumption change rate, the power factor, and the start-stop frequency into the dynamic Bayesian network, perform state inference based on the dynamic Bayesian network, and output the device start-stop state at the current moment; Among them, the device further includes: a threshold adjustment module, configured to dynamically adjust the threshold for judging the device start-stop state according to the changes in the device operation characteristics and the recognition results.

9. An equipment operation status identification device for sewage enterprises, characterized in that, It includes a memory and a processor, and the memory stores a computer program, and when the computer program is executed by the processor, it implements the method for identifying the operation state of the sewage discharge enterprise's equipment according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and when the computer program is executed by a processor, it implements the method for identifying the operation state of the sewage discharge enterprise's equipment according to any one of claims 1-7.

Images