JP5859866B2 - Monitoring target amount prediction method and monitoring target amount prediction apparatus - Google Patents

Description

  The present invention relates to a monitoring target amount prediction method and a monitoring target amount prediction apparatus for predicting a monitoring target amount of a plant facility such as a water treatment facility.

  In general, in water treatment facilities that perform sewage treatment (sewage), water purification (water supply), industrial wastewater treatment, etc., facilities to be monitored (for example, structures and equipment) and treatment conditions (for example, quality of treated water and sludge) The future state of a state etc. is predicted, and control object equipment is controlled based on a prediction result (for example, refer to patent documents 1). Specifically, the conventional monitoring target amount prediction method is suitable for the target water treatment facility because the fixed facility mainly predicts the future state of the water treatment facility that operates based on certain rules. The monitoring target amount is predicted using one prediction model using one prediction method, and the control target equipment is controlled based on the prediction result.

JP 2010-231457 A

  However, since the actual state of the water treatment facility changes from moment to moment, in the conventional monitoring target amount prediction method using one fixed prediction model, the state prediction is performed according to the timing of executing the state prediction. The accuracy of the water changes, and it becomes difficult to predict the state of the water treatment facility with high accuracy. For this reason, provision of the technique which can predict the state of water treatment equipment with high precision irrespective of the timing which performs state prediction was anticipated.

  The present invention has been made in view of the above problems, and its purpose is to provide a monitoring target amount prediction method and a monitoring target amount capable of predicting the monitoring target amount of plant equipment with high accuracy regardless of the timing of executing the prediction. It is to provide a prediction device.

  In order to solve the above problems and achieve the object, the monitoring target amount prediction method according to the present invention includes a plurality of different prediction models, plant facility operation result data, current operation state data, weather observation data, And a primary prediction step for calculating a plurality of primary prediction values of the monitoring target amount of plant equipment using the data relating to the weather forecast, and each primary prediction value predicted in the primary prediction step at the execution timing of the secondary prediction step. A secondary prediction step of assigning a corresponding weight and calculating a secondary prediction value of the monitoring target amount of the plant equipment as a prediction value of the monitoring target amount of the plant equipment using a plurality of primary prediction values to which the weights are given, It is characterized by including.

  In the monitoring target amount prediction method according to the present invention, in the above invention, the primary prediction step is based on plant facility operation result data, plant facility current operation state data, weather observation data, and weather forecast data. And regenerating or correcting a plurality of prediction models.

  The monitoring target amount prediction method according to the present invention is characterized in that, in the above invention, the plurality of prediction models include a neural network model, a multiple regression model, or a clustering model.

In the monitoring target amount prediction method according to the present invention, in the above invention, the secondary prediction step uses a neural network having a plurality of weighted primary prediction values as nodes, and uses a neural network having two nodes as the monitoring target amount of plant equipment. The method includes a step of calculating a next predicted value.

  The monitoring target amount prediction method according to the present invention is characterized in that, in the above invention, the secondary prediction step includes a step of calculating a predicted value of the monitoring target amount for each predetermined time within a predetermined period.

  In order to solve the above problems and achieve the object, a monitoring target amount prediction apparatus according to the present invention uses a plurality of different prediction models, operation result data of plant equipment, and data on current operation status, A primary prediction unit that calculates a plurality of primary prediction values of the monitoring target amount of equipment, a weight corresponding to the execution timing of the secondary prediction is assigned to each primary prediction value predicted by the primary prediction unit, and the weight is given Secondary prediction means for calculating a secondary predicted value of the monitored amount of plant equipment as a predicted value of the monitored amount of plant equipment using a plurality of primary predicted values.

  According to the monitoring target amount prediction method and the monitoring target amount prediction apparatus according to the present invention, it is possible to predict the monitoring target amount of plant equipment with high accuracy regardless of the timing of executing the prediction.

FIG. 1 is a schematic diagram illustrating a configuration example of a sewage treatment plant to which a monitoring target amount prediction system according to an embodiment of the present invention is applied. FIG. 2 is a block diagram illustrating a configuration of a monitoring target amount prediction system according to an embodiment of the present invention. FIG. 3 is a flowchart showing a flow of monitoring target amount prediction processing according to an embodiment of the present invention. FIG. 4 is a conceptual diagram for explaining the processing in steps S2 to S4 shown in FIG. FIG. 5 is a conceptual diagram for explaining a method of predicting a change in the monitoring target amount within a predetermined period. FIG. 6 is a conceptual diagram for explaining a method of predicting a change in the monitoring target amount within a predetermined period. FIG. 7 is a conceptual diagram for explaining a method of predicting a change in the monitoring target amount within a predetermined period. FIG. 8 is a conceptual diagram for explaining a method of predicting a change in the monitoring target amount within a predetermined period. FIG. 9 is a conceptual diagram for explaining the processing in step S5 shown in FIG.

  Hereinafter, a monitoring target amount prediction system according to an embodiment of the present invention will be described with reference to the drawings. In addition, although this embodiment applies this invention to the prediction process of the amount of monitoring objects used for the operation control of a sewage treatment plant, this invention is not limited to this embodiment, Sewage treatment The present invention can also be applied to prediction processing of monitoring target amounts used for operation control of plant facilities other than sewage treatment plants such as water treatment facilities for water purification treatment and industrial industrial wastewater treatment.

[Configuration of sewage treatment plant]
First, a configuration of a sewage treatment plant to which a monitoring target amount prediction system according to an embodiment of the present invention is applied will be described with reference to FIG. However, the configuration of the sewage treatment plant is not limited to the configuration shown in FIG.

  FIG. 1 is a schematic diagram illustrating a configuration example of a sewage treatment plant to which a monitoring target amount prediction system according to an embodiment of the present invention is applied. As shown in FIG. 1, a sewage treatment plant to which a monitoring target amount prediction system according to an embodiment of the present invention is applied treats sewage treatment equipment 1 that treats sewage and excess sludge generated in the sewage treatment equipment 1. Sludge treatment equipment 2 to be provided.

  The sewage treatment facility 1 includes a sand basin 11, a first sedimentation basin 12, a reaction tank 13, a final sedimentation basin 14, and a chemical injection tank 15. The sedimentation basin 11 is equipment for precipitating and removing non-septic inorganic substances such as earth and sand in sewage, and the sewage from which the non-septic inorganic substances have been removed is supplied to the first sedimentation basin 12 by a sewage pump 16. The first settling basin 12 is a facility for removing sedimentary organic substances in the sewage by sedimentation by gently flowing the sewage. First, the supernatant water of the sedimentation tank 12 is supplied to the reaction tank 13. The sediment in the initial sedimentation basin 12 is collected by the sludge scraper 17a and supplied to the sludge treatment facility 2 as excess sludge by the sludge extraction pump 18.

  The reaction tank 13 controls the amount of air supplied from the blower 19 to the reaction tank 13 according to indexes such as DO (dissolved oxygen concentration), MLSS (active sludge concentration in the tank), sewage flow rate, ORP (oxidation reduction potential). This is a facility that uses microorganisms to treat pollutants, mainly organic matter, nitrogen and phosphorus in sewage. The amount of air supplied to the reaction tank 13 is detected using a pressure gauge 20. The sewage from which the pollutants are removed is supplied to the final sedimentation basin 14, and a part of the sediment in the reaction tank 13 is supplied to the sludge treatment facility 2 as excess sludge by the sludge extraction pump 18.

  The final sedimentation basin 14 is a facility that separates the sewage supplied from the reaction tank 13 into treated water and activated sludge. The treated water is supplied to the chemical injection tank 15. The activated sludge is collected by the sludge scraper 17b and returned to the reaction tank 13 or supplied to the sludge treatment facility 2 as excess sludge by the sludge extraction pump 18. The chemical injection tank 15 is a facility for disinfecting the treated water by injecting chemicals into the treated water supplied from the final sedimentation basin 14 using the chemical injection pump 21 and then discharging the disinfected treated water to a river or the like. It is.

  The sludge treatment facility 2 includes a concentration tank 31, a digestion tank 32, a dehydrator 33, and an incinerator 34. The concentration tank 31 is a facility that concentrates excess sludge by removing moisture in the excess sludge supplied from the sludge extraction pump 18. Excess sludge concentrated in the concentration tank 31 is collected by a sludge scraper (not shown) and supplied to the digestion tank 32.

  The digestion tank 32 is a facility for decomposing organic matter in excess sludge to obtain digested sludge. Digested sludge generated in the digestion tank 32 is supplied to a dehydrator 33. The dehydrator 33 is a facility that reduces the moisture content of the digested sludge by dehydrating the digested sludge supplied from the digestion tank 32. The incinerator 34 is a facility that incinerates digested sludge whose water content has been reduced by the dehydrator 33.

[Configuration of monitoring target amount prediction system]
Next, the configuration of the monitoring target amount prediction system that is an embodiment of the present invention will be described with reference to FIG.

  FIG. 2 is a block diagram illustrating a configuration of a monitoring target amount prediction system according to an embodiment of the present invention. As shown in FIG. 2, the monitoring target amount prediction system 100 according to an embodiment of the present invention includes a performance data database 101, an input information acquisition device 102, a monitoring target amount prediction device 103, and an output device 104.

  The actual data database 101 stores actual values of various types of information used for monitoring target amount prediction processing as operation actual data. Specifically, the results database 101 includes weather (clear, clear, cloudy, rain (rain intensity)), rainfall, wind speed, amount of sunlight, water level of the treated water reservoir, and discharge of pumps such as the sewage pump 16. Actual values such as volume, amount of sewage inflow, amount of power demand, amount of generated power, amount of generated sludge gas, amount of generated sludge, etc. are stored as operation result data for each date and time. The operation result data is stored in the result data database 101 in a form that can be referred to by the monitoring target amount prediction apparatus 103 via the telecommunication line.

  The input information acquisition device 102 is connected to various devices in the sewage treatment plant, and includes weather, rainfall, wind speed, sunshine amount, treated water reservoir level, pump discharge amount, sewage inflow amount, power demand amount, generated power amount, Acquire current values such as sludge gas generation and sludge generation. The input information acquisition device 102 also acquires information regarding weather, rainfall, wind speed, and sunshine forecast. The input information acquisition device 102 outputs the acquired information to the monitoring target amount prediction device 103 as data relating to the current operation status of the sewage treatment plant.

  The monitoring target amount prediction device 103 is configured by a workstation such as a personal computer or an information processing device such as a process computer. The monitoring target amount prediction device 103 functions as a primary prediction unit 103a and a secondary prediction unit 103b when an arithmetic processing device such as a CPU in the information processing apparatus executes a computer program. The functions of these units will be described later.

  The output device 104 is for outputting various information from the monitoring target amount prediction device 103 to the control device, the display device, and the printing device of the sewage treatment plant.

[Monitoring amount prediction processing]
In the monitoring target amount prediction system 100 having such a configuration, the monitoring target amount prediction apparatus 103 predicts the monitoring target amount of the sewage treatment plant shown in FIG. 1 by executing the monitoring target amount prediction process shown below. Hereinafter, the operation of the monitoring target amount prediction apparatus 103 when executing the monitoring target amount prediction process will be described with reference to the flowchart shown in FIG.

  FIG. 3 is a flowchart showing a flow of monitoring target amount prediction processing according to an embodiment of the present invention. The flowchart shown in FIG. 3 is started every predetermined control cycle, and the monitoring target amount prediction process proceeds to the process of step S1.

  In the process of step S <b> 1, the primary prediction unit 103 a acquires the operation result data stored in the result data database 101 and data regarding the current operation status of the sewage treatment plant output from the input information acquisition device 102. Thereby, the process of step S1 is completed and the monitoring target amount prediction process proceeds to the process of step S2.

  In the process of step S2, the primary prediction unit 103a generates a plurality of primary models for predicting the monitoring target amount using the data acquired in the process of step S1. Specifically, as shown in FIG. 4, an example in which a neural network model, a multiple regression analysis model, and a clustering model are used as a primary prediction method, first, the primary prediction unit 103 a determines the characteristics of the sewage treatment plant as a prediction model. In order to reflect this, multi-variate analysis such as principal component analysis and correlation analysis, and start / stop of facilities and equipment for the operation result data (annual result data B and seasonal result data C) stored in the result data database 101 The reference prediction model A describing the relationship between the operation result data and the monitoring target amount is generated by analyzing the operation timing and the like. The reference prediction model A includes a reference model of a neural network model, a multiple regression analysis model, and a clustering model. In addition, the amount to be monitored includes power demand, power supply, sewage inflow, sludge gas generation, treated water quality (BOD, COD, turbidity, pH, ammonia content, etc.), and other objects that require monitoring. Amount included. However, the monitoring target amount can be arbitrarily selected and set.

  Next, the primary prediction unit 103a uses the annual performance data B and the standard prediction model A stored in the performance data database 101 to predict the annual average value of the monitoring target quantity. D is generated. The reference annual average model D includes an annual reference model of a neural network model, a multiple regression analysis model, and a clustering model as an example of FIG. In addition, the primary prediction unit 103a uses the seasonal result data C and the reference prediction model A stored in the result data database 101 to predict the average value of the monitoring target amount for each season. An average model E is generated. The reference seasonal average model E includes a neural network model, a multiple regression analysis model, and a clustering model seasonal reference model as an example of FIG.

  The processing up to this point may be executed in advance, and each model may be regenerated or corrected every predetermined period or every time the monitoring target amount prediction processing ends. When the equipment is started and stopped based on the predictive control, when the season changes and when abnormal weather (for example, guerrilla heavy rain) occurs, the operating state of the facilities constituting the sewage treatment plant may change suddenly. For this reason, the timing of executing prediction by comparing the state of the plant that changes from moment to moment with the state at the time of the previous prediction, and regenerating or correcting multiple prediction models when the difference exceeds the specified value Regardless of, the monitoring target amount of the sewage treatment plant can be predicted with higher accuracy.

  Next, based on the execution date and time of the monitoring target amount prediction process, the primary prediction unit 103a records the actual data of the current day and the previous day of the prediction execution, and the state of the sewage treatment plant on the day of the prediction run. A data group F is acquired which includes actual data on a day similar to the above and data on the power demand ratio on the day of the prediction execution calculated from the power supply-demand ratio prediction model. The demand ratio data refers to the power demand predicted from the relationship between the demand at a certain time on the forecast day (for example, 8:00 am) and the average demand for the same month, the same day, and the same time in the past several years. means. Then, the primary prediction unit 103a uses the acquired data group F, the reference annual average model D, and the reference seasonal average model E as an example of FIG. 4 as a neural model (applied method: neural network), multiple regression model ( Application method: multiple regression analysis), day before representative day model (application method: neural network and multiple regression analysis), representative day similar day model (application method: clustering model), and supply-demand ratio model (application method: clustering model) An annual average primary prediction model group G and a seasonal average primary prediction model group H are generated. That is, the primary prediction unit 103a regenerates or corrects a plurality of prediction models for predicting the monitoring target amount by using the operation result data of the plant facility and the data regarding the current operation state of the plant facility.

  The neural model is a model generated from the annual and seasonal reference models of the neural network model, and predicts the annual or seasonal average value of the monitoring target volume using a neural network whose operation data is a node. Model. The multiple regression model is a model that is generated from the annual and seasonal reference models of the multiple regression analysis model, and is a model that predicts the annual or seasonal average value of the monitored amount using the multiple regression analysis method. . The model on the day before the representative day is a model generated from the annual and seasonal reference models of the neural network model and the multiple regression analysis model, and is a model in which the operation result data on the day before the prediction execution is repaired and learned. The representative day similar day model is a model generated from the annual and seasonal reference model of the clustering model, and repairs the actual data of the day when the state of the sewage treatment plant is similar to the state of the sewage treatment plant on the day of the forecast execution It is a learned model. The supply-demand ratio model is a model generated from the yearly and seasonal reference models of the clustering model, and is a model in which the power supply-demand ratio data on the day of prediction execution is repaired and learned as described above. Thereby, the process of step S2 is completed and the monitoring target amount prediction process proceeds to the process of step S3.

  In the process of step S3, the primary prediction unit 103a uses the annual average primary prediction model group G and the seasonal average primary prediction model group H generated by the process of step S2 to calculate the prediction value of the monitoring target amount of each prediction model. Calculated as prediction result data J. In addition, as a prediction method of the monitoring target amount for 24 hours from a certain time, several methods as shown below can be illustrated. That is, when a neural model or multiple regression model is used, the primary prediction unit 103a predicts a monitoring target amount that is five minutes ahead from the present, and uses the predicted monitoring target amount, as shown in FIG. The monitoring target amount may be predicted at intervals of 5 minutes by predicting the monitoring target amount and repeating this operation 288 times. However, in this case, when the prediction accuracy of the monitoring target amount five minutes ahead is n, the prediction accuracy of the monitoring target amount 24 hours ahead is n to the power of 288, and the prediction accuracy gradually deteriorates.

  Further, as shown in FIG. 6, the primary prediction unit 103a predicts the monitoring target amount 24 hours ahead from the present, predicts the monitoring target amount 24 hours from the time after 5 minutes, and further performs this operation. By repeating 286 times, the monitoring target amount 24 hours ahead from the present may be predicted. In this case, 24 hours are required until the prediction process for obtaining the prediction results for 24 hours is completed, but the prediction accuracy of the monitoring target amount 24 hours ahead can be maintained. Further, as shown in FIG. 7, the primary prediction unit 103 a uses a certain prediction model to predict a monitoring target amount that is five minutes ahead from the present, and uses another prediction model to monitor a monitoring target amount that is ten minutes ahead. It is also possible to predict the monitoring target amount 24 hours ahead from the present by repeating this operation 286 times. In this case, however, 288 prediction models are required, and there is a possibility that the prediction accuracy varies. When the clustering model is used, the primary prediction unit 103a patterns the change in the monitoring target amount in the past 24 hours, selects a similar pattern from the current information, and gain (magnification), as shown in FIG. Alternatively, the monitoring target amount 24 hours ahead from the present may be predicted by the ratio (multiplier factor) and bias (additional constant). In this case, however, the number of patterns may become enormous. Thereby, the process of step S3 is completed and the monitoring target amount prediction process proceeds to the process of step S4.

  In the process of step S4, the secondary prediction unit 103b predicts the monitoring target amount using the prediction result data J calculated by the process of step S3 and the operation result data K stored in the result data database 101. A secondary prediction model is generated. Specifically, as shown in FIG. 4, first, the secondary prediction unit 103 b uses the operation result data (annual result data B and seasonal result data C) stored in the result data database 101. A reference secondary prediction model I is generated. The reference secondary prediction model I is a prediction model for a neural network that defines a correspondence relationship between operation result data and a monitoring target amount. Then, the secondary prediction unit 103b uses the prediction result data J calculated by the process of step S3 and the reference secondary prediction model I to predict the average value of the monitoring target amount for each year and season. A model L is generated. Thereby, the process of step S4 is completed and the monitoring target amount prediction process proceeds to the process of step S5.

  In the process of step S5, the secondary prediction unit 103b calculates the predicted value of the monitoring target amount using the secondary prediction model L generated by the process of step S4. Specifically, as shown in FIG. 9, the secondary prediction model 103 b is obtained by adding a secondary value to the predicted value of the monitoring target amount calculated using the annual average primary prediction model group G and the seasonal average primary prediction model group H. A weight corresponding to the execution timing of the prediction is assigned, and a prediction value of the monitoring target amount is calculated using a neural network having a plurality of monitoring target amounts to which the weight is assigned as nodes. Thereby, the process of step S5 is completed and a series of monitoring target amount prediction processes are completed.

  As is clear from the above description, in the monitoring target amount prediction process according to the embodiment of the present invention, the primary prediction unit 103a includes a plurality of different prediction models, operation result data of the sewage treatment plant, and current operation status. Data, meteorological observation data, and weather forecast data are used to calculate a plurality of primary prediction values of the monitoring target amount of the sewage treatment plant, and the secondary prediction unit 103b uses the secondary prediction step for each primary prediction value. Since weights are assigned according to the execution timing, the secondary predicted value of the monitoring target amount of the sewage treatment plant is calculated as the predicted value of the monitoring target amount of the sewage treatment plant using a plurality of primary predicted values to which the weights are given. Regardless of the timing of executing the prediction, the monitoring target amount of the sewage treatment plant can be predicted with high accuracy.

  Although the embodiment to which the invention made by the present inventor is applied has been described above, the present invention is not limited by the description and the drawings that form part of the disclosure of the present invention according to this embodiment. For example, in this embodiment, the primary prediction is performed using a neural network method, a multiple regression analysis method, and a clustering model method. However, the clustering model method is used to classify the actual data into small sets according to trends. After clarifying the trend, a prediction model suitable for each small set may be generated using a neural network method and a multiple regression analysis method, and primary prediction may be performed using the generated prediction model. Thereby, even if the past trend is difficult to grasp in the set of the entire record data, the tendency can be clarified and the primary prediction can be performed with high accuracy. As described above, other embodiments, examples, operation techniques, and the like made by those skilled in the art based on the present embodiment are all included in the scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Sewage treatment equipment 2 Sludge treatment equipment 11 Sedimentation basin 12 Initial sedimentation basin 13 Reaction tank 14 Final sedimentation basin 15 Chemical injection tank 16 Sewage pump 17a, 17b Sludge scraping machine 18 Sludge extraction pump 19 Blower 20 Pressure gauge 21 Chemical injection pump 31 Concentration tank 32 Digestion tank 33 Dehydrator 34 Incinerator 100 Monitoring target amount prediction system 101 Actual data database 102 Input information acquisition device 103 Monitoring target amount prediction device 103a Primary prediction unit 103b Secondary prediction unit 104 Output device

Claims (6)

Using multiple forecast models that differ from each other, plant facility operation results data, current operation status data, weather observation data, and weather forecast data, multiple primary prediction values for the plant facility monitoring target are calculated. A primary prediction step;
A weight corresponding to the execution timing of the secondary prediction step is given to each primary prediction value predicted in the primary prediction step, and a secondary of the monitoring target amount of the plant equipment using a plurality of primary prediction values given the weight A secondary prediction step for calculating the predicted value as the predicted value of the monitoring target amount of the plant equipment,
A monitoring target amount prediction method characterized by comprising:   The primary prediction step includes a step of regenerating or correcting a plurality of prediction models based on operation result data of plant equipment, data on current operation status of plant equipment, weather observation data, and data on weather forecast. The monitoring target amount prediction method according to claim 1.   The monitoring target amount prediction method according to claim 1, wherein the plurality of prediction models include a neural network model, a multiple regression model, or a clustering model. The secondary prediction step includes a step of calculating a secondary prediction value of a monitoring target amount of plant equipment using a neural network having a plurality of primary prediction values to which weights are assigned as nodes. The monitoring target amount prediction method according to any one of Items 1 to 3.   The monitoring target according to any one of claims 1 to 4, wherein the secondary prediction step includes a step of calculating a predicted value of the monitoring target amount every predetermined time within a predetermined period. Quantity prediction method. A primary prediction means for calculating a plurality of primary prediction values of the monitoring target amount of the plant equipment using a plurality of different prediction models, operation result data of the plant equipment and data on the current operation status;
A weight corresponding to the execution timing of the secondary prediction is given to each primary prediction value predicted by the primary prediction means, and a secondary prediction of the monitoring target amount of the plant equipment using a plurality of primary prediction values to which the weight is given. Secondary prediction means for calculating the value as a predicted value of the monitoring target amount of plant equipment,
A monitoring target amount prediction apparatus comprising:

Images